When you have a one-arm trial, you need a comparison group - the general problem presented by "External Control Arms."
Catch up with the newest thinking via a 3-arm webinar hosted by Friends of Cancer Research (streamed live on April 7, 2026).
Find it on YouTube here:
https://www.youtube.com/watch?v=fh7y3J2xyYY
Find the project home page at FoCR here:
https://friendsofcancerresearch.org/eca/
Below, Chat GPT summarizes the streaming auto-transcript.
###
On a personal note, I've seen several examples over the years where an entity used a one-arm real world study and an "external control arm" - 'administrative controls," patients in the same health plan with some similar diagnoses become "the control arm." And there can be various kinds of risk-balancing aka propensity score matching. Often the test group did so much better it was hard to attribute to the intervention and easier to attribute to a healthier population to start with. (This rarely gets buy-in from the inventor of the intervention). Personally, I found this the best explanation for the 2016 CMMI diabetes prevention project at YMCA's (Alva et al.) which used a propensity or diagnosis matched Medicare population. As I recall, even having 1 YMCA class was associated with big health gains in the following year. To me, this violated common sense dose-response expectations and left behind the suspicion that Medicare folk at YMCAs and tennis clubs might be healthither than the average beneficiary. But Alva et al. would be a "external control arm," as FoCR will discuss.
###
AI CORNER
##
Here is a detailed report on the Friends of Cancer Research webinar, “Application of External Control Arms in Oncology Drug Development,” held April 7, 2026.
Summary
This April 7, 2026 Friends of Cancer Research webinar showed a field moving from enthusiasm about external control arms to a more disciplined focus on fit-for-purpose data, prespecified methods, documentation, and early FDA engagement. The meeting’s core message was that ECAs are not a shortcut around rigorous evidence standards, but a practical tool when randomized trials are infeasible, especially in rare cancers. Speakers highlighted encouraging pilot work across multiple data partners, while stressing persistent challenges in missing data, endpoint alignment, population comparability, and regulatory predictability. Overall, the webinar suggested real progress in oncology ECA methods, but not a settled playbook for routine regulatory use.
###
Detailed report on the Webinar
This was not a “progress webinar” in the narrow sense of announcing a finished regulatory framework or a single definitive dataset. It was better understood as a field-building meeting: Friends used it to present interim results from its External Control Arm pilot, place those findings in the longer arc of its real-world-evidence work, and then test those findings against operational, clinical, patient, and policy perspectives. The central question running through the meeting was straightforward but consequential: when randomized controlled trials are infeasible or ethically difficult, how can external data be used rigorously enough to support oncology drug development and regulatory decisions? Friends framed the problem in exactly those terms, emphasizing rare cancers, small populations, and settings where control-arm randomization is hard to justify.
The opening keynote with Amy Abernethy set the intellectual frame for the rest of the event. Her main argument was that the field has moved well beyond the early question of whether real-world data can ever be used in regulatory science. In her telling, the more mature question is how to use it as part of “modern evidence generation”—that is, as one component of a broader evidentiary package that may also include prospective trials, with or without randomization. She described a clear technical evolution: the field has moved from claims alone, to claims plus EHR, then to PROs and genomics, and now toward longitudinal, multimodal, linked datasets that better represent the patient journey. Just as importantly, she argued that methodological maturity has also improved: more rigorous epidemiology, more attention to endpoint validity, better documentation, and greater FDA familiarity. But her subtext was equally important: the field is still organizationally fragmented, with trials, safety, epidemiology, engineering, and data infrastructure often living in separate silos. In that sense, the webinar’s message was not that ECA science is finished, but that it is finally being embedded into the larger machinery of drug development.
Abernethy also made an important use-case argument. She said the strongest immediate role for ECAs is in rare disease and rare cancer, where conventional randomization may be impractical, slow, or effectively impossible. But she did not stop there. She also pointed to ECAs as useful for early go/no-go development decisions, for post-approval or confirmatory contexts, and as a byproduct of prospective registries. That widened the frame considerably: ECAs were presented not as a niche workaround, but as a tool that could eventually support multiple moments in the product life cycle. Still, the keynote carefully avoided triumphalism. The recurring phrase was high-quality data, and the implication was unmistakable: the opportunity is real, but the evidentiary bar remains high.
The first substantive session focused on the Friends pilot itself. Bernat Navarro explained that the project generated six ECAs from different data sources, all attempting to reconstruct a control arm based on the control arm from a recently completed clinical trial, using a common protocol across independent data partners. This is an important design choice. Rather than asking whether one favored database can approximate a trial, Friends asked whether multiple heterogeneous sources, operating under a common plan, could do so in a reasonably coherent way. That is a more practical and more regulatorily relevant question, because real-world evidence programs rarely operate in a world of perfectly standardized data. Navarro also positioned this pilot as the latest step in a longer methodological series, following earlier Friends work on endpoint alignment, trial emulation, and real-world response.
A particularly strong aspect of the pilot was its choice of use case. The working group selected frontline metastatic pancreatic cancer, and the discussion makes clear why. It had regulatory relevance, multiple data partners could support it, an appropriate target trial existed, and the endpoint—overall survival—was comparatively robust and measurable across real-world and trial settings. Speakers explicitly contrasted overall survival with more fragile oncology endpoints such as response rate, duration of response, or progression-free survival, which often require RECIST-like operational definitions and more intensive harmonization. In other words, the pilot did not choose the most glamorous endpoint; it chose the most defensible one. That was consistent with the overall tone of the meeting: pragmatic, cautious, and method-first.
The headline methodological finding from Session 1 was encouraging but not simplistic. Navarro’s presentation suggested that independent construction of ECAs across heterogeneous external data sources is feasible under a shared statistical analysis plan. At the same time, the session repeatedly stressed that feasibility is not the same as interchangeability. Alignment with the target trial depended heavily on data fitness, including the availability of key prognostic variables, cohort size, and the ability to achieve baseline covariate balance. Matching or weighting could often improve comparability, but outcome estimates still varied across ECAs because data sources differ and methodological choices differ. The most revealing takeaway was probably not “ECAs work,” but rather “ECAs can be made to work better or worse depending on data completeness, cohort construction, and analytical discipline.” That is a more sober, and probably more useful, conclusion.
The panel discussion following that presentation sharpened the realism. Several speakers emphasized that clinical trial patients are often systematically different from real-world patients, especially in an aggressive disease such as metastatic pancreatic cancer. One panelist noted that trial participation itself requires the “luxury of time” to undergo screening, a built-in selection mechanism that tends to favor fitter patients. That is a crucial point, because it means ECAs are never just about reproducing trial eligibility criteria on paper; they are also about confronting the fact that trials and practice may draw from different patient universes even before analysis begins. The discussion of race was also notable: one speaker observed that data sets whose racial mix aligned more closely with the trial appeared to perform more similarly, suggesting that demographic comparability may matter alongside clinical comparability. The meeting did not oversell that observation, but it flagged it as a serious issue.
The webinar spent substantial time on missingness, curation, and data provenance, and that was one of its strongest sections. Mari Bradley from FDA said plainly that there are no simple thresholds for “how much missingness is acceptable”; acceptability depends on the intended regulatory use and the clinical context. Neil Maripole added a practical corollary: sponsors need to understand not just who is in a data source, but how the data were curated—for example, whether key variables came from human abstraction, machine learning, or LLM-based extraction, and how each performed against a reference standard. This was one of the meeting’s most important messages. The real regulatory question is not “Do you have data?” but “Can you defend how those data came into existence, how reliable they are, and whether the endpoint can be measured consistently across arms?”
Session 2 translated those lessons into drug development strategy. Jacquellyn Bosco framed the entire conversation with a simple but useful distinction: ECAs can be used descriptively, contextually, or with true regulatory intent, and those uses should not be blurred. Once a sponsor is thinking about regulatory use, she argued, ECA planning has to start up front, not after the trial is over. She warned against overreliance on simple feasibility counts and stressed the need for more serious quantitative feasibility work, including attrition and longitudinal missingness. That point was echoed by others, who argued that sponsors should begin designing trials with real-world comparability in mind rather than trying to retrofit an ECA later. This was one of the webinar’s clearest practical messages for industry: if ECAs are part of the eventual evidentiary strategy, they must be built into the development plan from the beginning.
The patient and advocacy dimension was not treated as decoration. Patient advocates emphasized that when advocacy groups or disease communities build real-world datasets, they should think prospectively about consent, data access, and downstream reuse, including possible ECA applications. There was also a repeated tension between patient-centered innovation and historical data availability. Amy Comstock Rick captured that especially well in the policy session: there is increasing pressure to use endpoints that reflect patient preferences, but some of those outcomes were not collected in older data sources, creating a conflict between being innovative now and remaining comparable to the historical record. That is a subtle but important point. The future of ECA work may depend as much on how today’s data are collected for tomorrow’s questions as on how legacy data are analyzed today.
The policy panel was perhaps the most candid part of the event. It did not claim that the evidentiary expectations for ECAs are fully settled. Mark Lee drew a distinction between two settings. For early development, expectations are often clearer because the sponsor sets the decision threshold for internal phase-transition purposes. For regulatory applications, however, speakers suggested the field is still not fully at a place of predictability. Amy Comstock Rick said the standard of substantial evidence of effectiveness has not changed, even if interest in mechanistic data and ECAs is growing. Joe Franklin’s contribution reinforced the industry reality: sponsors face a real investment question about whether to spend time and money pursuing FDA engagement around an ECA strategy, which in turn makes early and intensive communication with FDA critical. Donna Rivera added a useful organizing device, describing a “six C’s” problem set that includes complex design, completeness of capture, comparability of populations, and confounding/bias, all under the umbrella of regulatory alignment. The common thread across these remarks was that the barrier is not a lack of imagination; it is the difficulty of making ECA evidence predictable, transparent, and reviewable at regulatory grade.
At the same time, the policy panel was not pessimistic. One FDA speaker, discussing very small rare-disease populations, said that “that’s infeasible” cannot be the end of the conversation. That line nicely captured the meeting’s mood. The bar is not being lowered, but the system is increasingly willing to consider alternatives where conventional RCT structure fails. Mark Lee pushed that further by arguing that the long-term aspiration is broader than rare disease alone: ECAs should eventually help run smaller, more efficient studies and bring some of that same efficiency to the regulatory side, without relaxing scientific standards. The phrase that best summarizes the meeting may be this: not a lower bar, but a better method for clearing a high bar in situations where the old route no longer works well.
One especially forward-looking part of the webinar was the discussion of prospective ECAs. Mark Lee argued strongly that sponsors should increasingly consider building ECAs intentionally and prospectively, especially in settings where patients may no longer accept randomization easily, or where post-approval crossover contaminates overall-survival analyses. He even sketched the possibility of a built-in validation strategy in which an external control could first be benchmarked against an internal control on one endpoint before being “ungated” for another analysis such as overall survival. Whether that exact architecture proves practical is another matter, but it shows where the more ambitious part of the field is heading: away from ad hoc retrospective rescue missions and toward planned external-control infrastructure.
Bottom line
My reading of this webinar is that it showed real progress, but not resolution. Friends of Cancer Research and its collaborators have moved the ECA conversation past general enthusiasm and into the harder zone of comparability, missingness, endpoint selection, documentation, and regulatory process design. The pilot results suggest that heterogeneous data sources can, under a common plan, produce externally controlled comparisons that are at least analytically serious. But the meeting was equally clear that success depends on data provenance, prespecification, fit-for-purpose endpoints, and early regulatory strategy. The webinar therefore marked a step forward not because it “proved” ECAs are ready for routine approval use in oncology, but because it showed the field is learning how to ask the right questions in a disciplined way.
If you want,
