Something has shifted in the room.
The recent GetReal Conference 2026 in Utrecht, brought together regulators, health technology assessment (HTA) bodies, payers, academia, and life science partners, and the conversation about real-world evidence (RWE) felt different. The question is no longer whether RWE belongs in the evidence ecosystem. It is now about when to use it, how to use it well, and which type of evidence is the right fit for the question at hand. That shift, from skepticism to strategic integration, was palpable throughout every session.
Here is what we heard, and what it means for the future of evidence generation in oncology.
The conference opened with a keynote by Carole Longson, former Executive Director at NICE, who pioneered NICE’s health technology evaluation approach. In a bold statement, Dr. Longson called for a fundamental rethinking of how we classify and evaluate evidence. The traditional evidence pyramid, with its strict hierarchy placing randomized controlled trials (RCTs) explicitly above observational cohort studies, was designed for a different era. Today, observational study methods have advanced significantly and high-quality real-world data sources have become increasingly available. In addition, over the past decade, we have seen a dramatic shift in clinical innovation: therapies are adopted faster, disease cohorts are often biomarker-specific and thus smaller, and the pace of therapeutic development outstrips the timelines of traditional evidence generation.
Longson's argument was not that RCTs are less valuable. It was that our attachment to the hierarchy is preventing us from asking the right questions. Labeling evidence by its position in a hierarchy can obscure its actual fitness for purpose. What we need instead is a framework that asks: what is the right evidence to answer this specific question, for this specific population, at this specific point in a therapy's lifecycle?
Longson also pointed to two concrete examples of RCT and RWE working together: augmenting post-approval trial evidence with prospectively collected real-world data, and using synthetic or external control arms for single-arm trials. The key to success in both cases, she emphasized, is rigorous study design. Poorly conducted observational research does not just fail, it sets back the field.
These ideas resonated throughout the conference. It is a framing we find compelling at Flatiron, and one we are actively working to embed in how we approach evidence generation with our partners. To put this framework into practice, we commonly support clients by deploying advanced methodologies to evaluate the transportability of real-world study results between different geographies. Whether through the rigorous application of external control arms for (single-arm) trials or comprehensive clinical trial contextualization, our goal is to deliver a fit-for-purpose body of evidence that robustly addresses the specific questions of international decision-makers.
A consistent theme across sessions was the gap between what evidence is generated and what decision-makers can actually use. One panel speaker affiliated with the European Medicines Agency was direct: regulators need data they can trust, understand, and know is fit for purpose. That means data quality, analytical transparency, and reproducibility are non-negotiable.
In that same panel discussion, a different point was made that deserves wider attention: RWE could be applied across all elements of the PICO framework (population, intervention, comparator, and outcome), yet in practice it is almost exclusively used to characterize patient populations. This is an untapped opportunity for the RWE community. For rarer diseases and smaller patient cohorts, where randomized trials are either not feasible or not ethical, RWE is being underutilised in addressing the 'I', 'C', and particularly the 'O': whether it is the previously mentioned application of real-world external control arms to stand in for absent comparators, characterizing the natural history and real-world outcomes to contextualize (single-arm) trial results, or the demonstration of what meaningful clinical benefits actually look like in routine practice. Done well, this type of evidence can move RWE from a descriptive tool into a genuine driver of access decisions for the patients who need it most.
The challenge is not just generating high-quality RWE, but communicating its value clearly to the people who need to evaluate it. HTA assessment committees, in particular, need support in understanding why and how non-traditional evidence types are valid and useful. Bridging that communication gap is as important as the science itself.
The EU Joint Clinical Assessment (JCA) dominated much of the conference, and for good reason. The message from assessors was unambiguous: empty chapters in a dossier are not acceptable without explanation. Developers are expected to demonstrate that they have explored all potential evidence options, justify what they chose not to include, and explain any gaps. An evidence gap that exists because data is genuinely unavailable may be accepted, but only if that case is made explicitly and rigorously.
Transparency is equally critical for methods. Using a conference abstract as the basis for an indirect comparison, with no details on how outcomes were calculated, will not meet the bar. Full analytical details are required.
The JCA is still a learning exercise and both assessors and developers are adapting as they go, but the direction is clear. RWE will play a growing role as developers begin building evidence strategies earlier, and as national HTA bodies increasingly look to real-world data to supplement JCA outputs for local reimbursement decisions.
There is a meaningful opportunity here for the utilization of Flatiron’s research databases, as their clinical depth enables robust, patient-level analyses across all four PICO components. Flatiron’s data and analytical capabilities are well positioned to plan for and generate the transparent, reproducible analyses that JCA assessors require, well ahead of regulatory submission timelines.
In the discussion session on federated data networks it was pointed out that these platforms do not make underlying data better, rather they help researchers understand what data is available to them and if it is fit to answer a specific question at hand. The harder, more important work is improving data at source as true quality relies heavily on the completeness of data points. This is the gap Flatiron strives to fill. In the US, UK, Germany, and Japan, Flatiron is building fit-for-purpose, cancer-specific datasets curated from source EHR data, combining structured and unstructured records, expert clinical abstraction, and AI-powered natural language processing to achieve high completeness across hundreds of unique, standardised variables, all underpinned by tumor type-specific common data models which enable rigorous multinational research.
AI also featured prominently in the discussions, particularly with respect to how and where it is appropriate to be used. There appeared to be comfort with the use of large language models (LLMs) to support the execution of synthesis tasks such as systematic literature reviews. But what role, if any, should it play in the generation of data, or the execution of analyses? And what level of human oversight is required for decision makers to feel comfortable trusting the output? The question the field is grappling with is not whether it is technically feasible to use this evolving technology to enhance evidence generation, but under what conditions is it appropriate and trustworthy?
These are exactly the kinds of methodological questions that require close collaboration between data providers, life science partners, and regulators and the creation of peer-reviewed frameworks that establish a baseline of credibility. One example is Flatiron’s VALID (Validation of Accuracy for LLM/ML-Extracted Information and Data) framework which was discussed at the conference. It is a practical, transparent, and holistic framework for evaluating the quality and fitness-for-purpose of real-world data (RWD) extracted by machine learning (ML) and LLMs from electronic health records (EHRs). Published in the Journal of Clinical Oncology Clinical Cancer Informatics in April 2026, it is recognized as the first comprehensive, peer-reviewed validation framework for AI-extracted real-world oncology data in the industry. Importantly, the framework aims to improve transparency of AI-data curation processes through a comprehensive approach: benchmarking model extractions directly against expert human abstraction, running automated clinical consistency and plausibility checks, and conducting independent replication analyses of key outcomes. Data holders should ensure to have appropriate frameworks in place to demonstrate their level of clinical accuracy and thus instill trust in their extraction processes and ultimately their data products.
Taken together, the conversations at this conference point to a field in transition. The question is no longer whether RWE has a role. The question is how to use it well, how to communicate its value, and how to build it into evidence strategies from the start, not as an afterthought.
At Flatiron, we have spent over a decade building the data infrastructure, scientific expertise, and regulatory relationships to support exactly this kind of evidence generation. Whether you are thinking about an EU JCA submission, an HTA dossier, a regulatory filing, or an integrated evidence plan that spans all three, we would welcome the conversation.
Want to go deeper on these topics? Join us on July 28 at 3PM BST / 4PM CEST for our free webinar: “An HEOR playbook for oncology RWE and AI use to support European HTA”. We’ll explore how to put these insights into practice.