Ask a medical-affairs lead how they intend to build the real-world-evidence case for label expansion, and you tend to get one of two answers. We will stand up a registry. Or: we will run one good regulatory-grade study and take it to the agency. Both sound like the responsible move, and both quietly bet everything on the wrong stage of a longer game.
Here is the number that should give you pause. Deng, Girman and Ritchey went through 218 labelling-expansion approvals between January 2022 and May 2024 and asked how often real-world evidence featured. Roughly a quarter of the time, year on year: 23.3% in 2022, 27.7% in 2023, 23.7% across the slice of 2024 they captured [1]. Then they looked at what kind of RWE actually showed up in the expansions that succeeded. Electronic health records accounted for 75% of it; registries, for 4.6% [1].
Read that again: the data source almost everyone reaches for first is the one that least often closes the case.
So let me put the argument plainly. A case for a new indication is a sequence, not a single study. Every data source you own belongs to a particular stage of that sequence, fit for one job and useless for the next, and the whole thing rises or falls on how you order it. The real question is which stage you're standing on, and what that stage owes the next one, not which study to run.
Think of a supplemental submission as an argument made in front of a hostile examiner. You cannot open with the conclusion, that the drug works in the new population, and expect it to land. You have to lay the premises first: here is what this disease does when you leave it alone, here is a signal that the drug bends that course, here is that signal confirmed to the standard the examiner will actually hold your final claim to. Skip a premise and you are asserting the verdict before the proof. The examiner always notices.
Three stages, then: size it, signal it, settle it.
Be honest about the provenance of that three-part frame: no single guidance document or paper sets it out as its own named architecture for label expansion. It is a synthesis. The clearest agency-sourced version of it is the EMA's Guideline on registry-based studies, which splits a registry's usefulness cleanly across the lifecycle. Pre-authorisation, a registry can supply "incidence, prevalence and determinants of disease outcomes" and help "contextualise the results of uncontrolled trials" [3]. Post-authorisation, registries "can be the basis for recruitment and randomisation for RCTs and non-interventional studies, post-authorisation efficacy studies (PAES) and post-authorisation safety studies (PASS)" [3]. Note the honest gap: that guideline frames the second act around PAES and PASS, not around a sponsor-initiated push for a new indication. The staging idea is theirs; the label-expansion application is ours.
Abstractions are cheap, so here is a sequence you can actually name, trace and pull up on ClinicalTrials.gov. Eculizumab (Soliris) was approved for paroxysmal nocturnal haemoglobinuria in 2007, then expanded into atypical haemolytic uraemic syndrome in 2011, and the way the evidence was staged is worth watching.
A caveat. ClinicalTrials.gov records what a study measured, not what a regulator did with it. That these specific trials underpinned the 2011 aHUS approval is the public regulatory history, not a line item in the registry record. But the shape is rather clear: a hypothesis-generating chart review that hands off to a pivotal, and the pivotal to a registry. Honestly, this is one of the most elegant CDPs I've seen.
Free download
The RWE Briefing Document Template
The section-by-section structure for the RWE part of a regulatory briefing, built around the questions reviewers actually ask.
Get the template →So what is a registry actually good for, then? A registry is a superb instrument for sizing a disease and a poor one for settling a claim. That's the entire point of staging, not a knock on registries. The evidentiary bar climbs as you move through the sequence, and each rung reorders which data source is fit to stand on it.
The Deng breakdown is the mechanism in miniature. Across those successful labelling expansions, cohort designs made up 87.5% of the RWE and 65.9% were retrospective, while registries supplied that lonely 4.6% [1]. Registries dominate the sizing conversation, and when it comes to the confirmatory one they almost never clear it. The EMA says the same thing in methodological language in its 2025 Reflection paper on real-world data in non-interventional studies, which draws a hard line between a study "designed to describe patient characteristics without regards to any causal hypothesis" and one "designed to investigate the effect... in comparison to what would have happened... under non-exposure" [4]. Descriptive and causal are different jobs, and a data source can be excellent at the first and hopeless at the second.
Ivacaftor (Kalydeco) shows the fitness question with unusual clarity. It went from the G551D mutation to the broader set of non-G551D gating mutations, and its first stage in that new population was NCT01614470, a Phase 3 randomised, placebo-controlled crossover trial in 39 patients, not real-world evidence at all. The stages are not all "real-world" by definition; you pick the design the bar demands and the population allows. Only later came NCT02445053 (VOCAL), a prospective, 48-month, single-arm observational cohort of 75 patients measuring the hard, long-horizon outcomes that a short crossover trial simply cannot see, mortality, transplantation, four-year lung-function trajectory. That single-arm real-world design was credible precisely because the randomised trial had already nailed the mechanism-linked effect. On its own, at stage one, it would not have carried the expansion.
This is where the reflex to trust "RWE that matched the RCT" needs a caveat. The RCT-DUPLICATE programme emulated 10 trials with real-world data and reached regulatory agreement in 6 of them, with the RWE hazard ratio landing inside the trial's 95% confidence interval in 8 of 10 [8]. The extended 32-trial follow-up found an overall correlation of 0.82, rising to 0.93 in the subset that emulated the original trial design most closely [9]. The lesson is not "RWE agrees with RCTs." Agreement is conditional on how well you emulate, and it is not a property RWE carries around by category. So the Monday-morning move is simple to say and hard to do: choose the data source by the stage, and accept that a claims extract perfect for sizing is not fit for settlement. If you want the fuller argument on what earns a data source that word, we have written on what "regulatory-grade RWE" actually means and on the EMA's own thinking on single-arm designs like the eculizumab pivotals.
So where does a real-world-evidence sequence for label expansion actually break?
There are exactly two failure modes, and they are opposites.
The skip. You jump to a confirmatory-grade study without the stage-one and stage-two groundwork that would have told you whether a signal was even there. What that looks like when it reaches the agency under-built is documented. If you want the regulator's-eye view of that same failure, we've covered how to use RWE to support regulatory strategy elsewhere. Alipour-Haris and colleagues examined 85 pre-approval RWE use cases across the FDA and EMA between 2016 and 2022; 24 of them (28.2%) were for label expansion, and 13 across the set were rated "not supportive," for a recurring cast of reasons that included small sample size, selection bias, missing data, misclassification and confounding [7]. Worth sitting with: in 76.5% of the use cases where RWE served as the primary evidence, it did so with no new clinical trial alongside it [7]. When that is your whole hand and a stage was skipped, those failure reasons are what the reviewer writes down.
The stall. This one is quieter and, in its way, more wasteful. Call it the "standing registry": an evidence-generation machine that measures forever and never graduates a discrete, dated campaign. The ICGG Gaucher Disease Registry (NCT00358943) has run since 1991, across some 60 countries, toward a 12,000-patient target, explicitly tasked to "evaluate the long-term effectiveness of imiglucerase and of eliglustat" [5]: thirty-five years of real, careful data. Yet its record shows no distinct, dated comparative study that converted all that observation into a specific, nameable label change. I want to be precise here, because the evidence is: this is not a claim that the registry failed, only that the onward sequence is not visible in the record. It is a survey crew that has mapped the same ground for three decades and never once broken ground on the building.
One honest limit on all of this. There is no published attrition rate for stalled label-expansion programmes, because every denominator in the literature is drawn from completed, public approval packages, and a programme that quietly never files leaves no trace to count. So do not let anyone sell you Arondekar's 0.8% or Alipour-Haris's 13 unsupportive cases as a stall rate; they are not one. The stall is diagnosable in your own programme long before it is ever measurable across the industry.
A lean biotech with eighteen months of runway cannot fund a three-study arc toward a supplemental indication, so the rational move is the single study the regulator will accept, and go. And the numbers seem to back the sceptic: Arondekar and colleagues found that of 573 supplemental oncology approvals between 2015 and 2020, 249 were for new indications, and only two of those, 0.8%, used RWE in support of efficacy, against 8.3% of original oncology approvals in the same window [6]. If RWE almost never carries an oncology expansion anyway, why spend a penny sequencing it?
Because the sequence is not three expensive studies, and that is the misread. The early stages are the cheap ones: existing epidemiology you can license, a retrospective chart review, a feasibility assessment done before anyone writes a protocol. Their entire job is to tell you whether the one expensive confirmatory study is worth running at all. Skip them and you do not save runway, you move the failure later, to the point where it costs the most and teaches the least.
The 0.8% is the argument for sequencing, not against it. RWE clears the confirmatory bar rarely, so you spend it where it is genuinely fit, sizing, signal-finding, contextualising an uncontrolled trial, and you reserve the settling stage for the design the bar actually demands, which may be RWE or, as with ivacaftor, a small randomised trial. One caution on the numbers, since I would rather you did not overclaim on my account: Arondekar's 0.8% (oncology, 2015–2020) and Deng's 23–28% (all therapeutic areas, 2022–2024) are not one trend line. Different areas, different eras, different denominators. Real growth is plausible; a clean "from under 1% to over 20%" story is not something either dataset can support alone.
This is the same discipline as picking your critical few evidence gaps in the 90/10 minimum viable evidence framework: every stage has its own definition of "good enough," and spending stage-three money on a stage-one question is how lean programmes bleed out. It is also, more or less, the entire case for running an integrated evidence plan when resources are tight. The sequence is exactly what an IEP exists to hold together, the living document that makes evidence graduate from one stage to the next instead of quietly accumulating in a registry no one has a plan for.
There is a small irony worth ending on. If you want the FDA's thinking on how to build stage one properly, the natural-history baseline, the document you reach for is "Rare Diseases: Natural History Studies for Drug Development," and it has been sitting at DRAFT since March 2019, seven years and counting. The one piece of guidance most directly about how to build the first stage of the sequence has itself never cleared its own first stage.
It is emblematic of a wider gap: there is no FDA guidance titled or scoped specifically for using RWE to support an efficacy supplement. The Framework for FDA's RWE Program (December 2018) names label expansion as a statutory purpose of the programme under §505F, and the "Considerations for the Use of Real-World Data and Real-World Evidence" guidance (FINAL, August 2023) is the closest thing on the shelf, but both are cross-cutting and leave the staging for you to construct.
So construct it deliberately: name your stages, know which one you are on, and know what it owes the next. The single study was never the unit of the case. The sequence was.
Get the monthly digest
The 5 things evidence leads need to know each month: regulatory moves, RWE developments and what they mean in practice. No pitch, one email a month.
[1] Deng YF, Girman CJ, Ritchey ME. (2025). "Real-World Evidence in FDA Approvals for Labeling Expansion of Small Molecules and Biologics." Therapeutic Innovation & Regulatory Science;59(5). PMID: 40468096. PMC12446098. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12446098/
[2] Berger ML, Sox H, Willke RJ, et al. (2017). "Good practices for real-world data studies of treatment and/or comparative effectiveness: Recommendations from the joint ISPOR-ISPE Special Task Force on real-world evidence in health care decision making." Pharmacoepidemiology and Drug Safety;26(9):1033–1039. PMID: 28913966. https://pubmed.ncbi.nlm.nih.gov/28913966/
[3] European Medicines Agency. "Guideline on registry-based studies." EMA/426390/2021. FINAL, adopted 16 September 2021.
[4] European Medicines Agency. "Reflection paper on the use of real-world data in non-interventional studies to generate real-world evidence." EMA/99865/2025. FINAL, adopted 17 March 2025.
[5] Sanofi Genzyme. "International Collaborative Gaucher Group (ICGG) Gaucher Registry." ClinicalTrials.gov: NCT00358943. https://clinicaltrials.gov/study/NCT00358943
[6] Arondekar B, Duh MS, Bhak RH, et al. (2022). "Real-world evidence in support of oncology product registration: A systematic review of new drug application and biologics license application approvals from 2015-2020." Clinical Cancer Research;28(1):27–35. PMID: 34667027. https://pubmed.ncbi.nlm.nih.gov/34667027/
[7] Alipour-Haris G, et al. (2024). "Real-world evidence to support regulatory submissions: A landscape review and assessment of use cases." Clinical and Translational Science;17(8). PMID: 39092896. https://pubmed.ncbi.nlm.nih.gov/39092896/
[8] Franklin JM, Patorno E, Desai RJ, et al. (2021). "Emulating randomized clinical trials with nonrandomized real-world evidence studies: First results from the RCT-DUPLICATE initiative." Circulation;143(10):1002–1013. PMID: 33327727. https://pubmed.ncbi.nlm.nih.gov/33327727/
[9] Wang SV, Schneeweiss S, et al. (2023). "Emulation of Randomized Clinical Trials With Nonrandomized Database Analyses: Results of 32 Clinical Trials." JAMA;329(16). PMID: 37097356. https://pubmed.ncbi.nlm.nih.gov/37097356/
Trials cited inline: eculizumab sequence — NCT01770951, NCT00838513, NCT00844545, NCT01522170, NCT01522183 (https://clinicaltrials.gov/study/NCT01770951 and siblings); ivacaftor sequence — NCT01614470, NCT02445053 (https://clinicaltrials.gov/study/NCT01614470, https://clinicaltrials.gov/study/NCT02445053).
FDA guidance cited by title/date/status only: Framework for FDA's Real-World Evidence Program (December 2018); Considerations for the Use of Real-World Data and Real-World Evidence to Support Regulatory Decision-Making for Drug and Biological Products (FINAL, August 2023); Rare Diseases: Natural History Studies for Drug Development (DRAFT, March 2019). Title, date and DRAFT/FINAL status independently confirmed via the Federal Register and the HHS Guidance Portal; no FDA body text is quoted.