From RWE Policy to Practice: Building Evidence Decision-Makers Can Trust
Reflections from the 2026 Duke-Margolis Convening on the State of Real-World Evidence Policy
By Ingeborg Holt | July 29, 2026
Key takeaway: RWD becomes decision-ready only when its fitness is evaluated against a defined question, intended use, and evidentiary role.
The Duke-Margolis Institute for Health Policy held its annual State of Real-World Evidence Policy convening on Monday, July 27, 2026. The discussion brought together leaders from government, health systems, industry, academia, and data organizations to examine how real-world data and real-world evidence (RWD/E) can more routinely inform regulatory, coverage, payment, and care-delivery decisions.
My strongest takeaway was simple: the field is no longer asking whether RWE has value. The harder, and more important, question is how to generate evidence that is sufficiently relevant, reliable, transparent, and decision-ready for its intended use. This is no small issue. As a recent scientific publication noted, “Despite a decade of investments and regulatory enthusiasm, real-world evidence (RWE) generated from real-world data (RWD) remains underutilized in drug approvals” (Fakhouri, 2026).
The opening session also highlighted new developments at the Centers for Medicare & Medicaid Services (CMS) and FDA’s Center for Devices and Radiological Health (CDRH), including the CMS Innovation Center’s ACCESS model and CDRH’s TEMPO pilot. Although both warrant a separate discussion, this post focuses on the implications of RWD policy for drug and biologic programs overseen by CDER and CBER.
Start with the question, not the dataset
Across the sessions, speakers repeatedly returned to the need to begin with the question to be answered and then evaluate the available evidence sources. For premarket regulatory use, that means defining the regulatory objective and the role RWE will play in an application: pivotal evidence, supportive evidence, contextual information, or natural history.
This distinction matters because no real-world dataset is universally “high quality” or fit for every purpose. Claims data may be well suited to characterize health care utilization or indicate prescription use (or at least prescription pickup) but may be insufficient for an endpoint that depends on detailed clinical assessments. Electronic health record data may contain clinical richness while still presenting challenges involving missingness, inconsistent documentation, fragmented longitudinal histories, or variable endpoint capture. Registries may offer focused clinical depth but have limited representativeness or comparators. This is why FDA does not provide a single checklist for using RWE and instead provides guidance and encourages drug developers to engage the Agency early when considering RWD sources and study designs.
A fit-for-use assessment of RWD cannot be a generic scorecard completed after a data source has been selected. Fitness for purpose is not an inherent characteristic of a real-world dataset; it emerges from the alignment between the data, their intended role, and the research question. The assessment must connect the clinical question with the target population, exposure, outcomes, and timing to the actual properties and limitations of the data.
AI expands what RWD can reveal—and what must be evaluated
The session on AI and RWE explored the opportunity to use AI to extract information that has historically been inaccessible (or accessible only through substantial manual effort) from unstructured data such as clinical notes. AI may help reconstruct longitudinal patient journeys, identify the rationale behind clinical decisions contained in written notes, support curation at scale, and make data more reusable.
However, AI does not eliminate data-quality problems; it adds new dimensions to them. Decision-makers need to know which fields were AI-derived, what reference standard was used, how outputs trace back to source records, how performance varies across populations, and whether results remain reproducible as models and workflows change. Validation must address not only extraction accuracy for individual variables, but also whether variables make clinical sense together and whether analyses based on AI-curated data yield trustworthy results. AI must also be continually revalidated to prevent drift and remain current with coding conventions, evolving standards of care, and other clinical practices.
Documentation, provenance, versioning, human oversight, and ongoing performance monitoring must therefore be built into the evidence lifecycle when AI is used. These controls are not obstacles to innovation; they make responsible scaling possible. Several panelists noted that comparing AI performance with human performance can prevent unrealistic expectations of perfection. That context is important, but AI deployed at scale can propagate errors far more rapidly than an individual human process. As with other public health technologies, the potential benefits must be weighed against the risks.
RWE must become part of the learning cycle
The learning health system discussion made clear that data alone do not produce learning. Organizations also need timely data access, interoperable infrastructure, workflows that can act on findings, a workforce able to translate analysis into action, sustained leadership support, governance, and reliable funding. As one panelist summarized it: “the will, skill, and dollar bill.”
A mature evidence ecosystem needs an intentional cycle in which new evidence is generated, assessed, communicated, and used to revisit decisions. This requires infrastructure and the ability to standardize data and make them interoperable. Panelists also noted that negotiating data-sharing agreements needed to integrate data into a learning health system takes considerable effort and that a standardized or streamlined process to facilitate these agreements would be tremendously helpful.
Fitness checks should guide a proceed, adapt, or reassess decision
The final regulatory policy session, moderated by Motiur Rahman of CDER’s Office of Medical Policy, offered one of the convening’s most practical messages: conduct an integrated fitness check before making costly commitments. He acknowledged industry concerns about regulatory uncertainty but emphasized that challenges often arise when sponsors do not engage FDA before conducting their study. FDA has consistently recommended in public meetings and in guidance that sponsors planning to use RWD in a submission meet with the Agency to discuss data-source selection, the study protocol, and the statistical analysis plan before conducting the study.
Fit-for-use evaluation of real-world data sources should function as a decision gate, not a documentation exercise. Asking the right questions about the data creates an opportunity to identify issues early: Are eligibility criteria routinely captured? Are important confounders available? Is missingness likely to affect interpretation? Is follow-up time sufficient? Is the study population representative of the target population? Can submitted data be traced to the original source? Can patient-level data be submitted to FDA? The answers can support a decision to proceed, adapt the approach, or reassess whether to use the data.
Turning real-world data into decision-ready evidence
The objective is not to label a dataset “fit” or “unfit.” It is to determine, transparently and defensibly, whether the data can answer the question well enough to support the decision at hand.
Organizations that evaluate RWD fitness for use early and effectively by asking the right questions will be best positioned to choose real-world data that can generate trustworthy evidence.
Orizaba Solutions is ready to help assess your data for regulatory decision-making. Our assessment includes questions across five domains: Study (or Intended Use) and Data Source Profile; Prespecification; Data; Feasibility; and Privacy and Regulatory. Questions are grounded in FDA guidance, subject-matter expertise, and reviews of RWE-supported FDA submissions.
Explore our Health IT services and learn more about our RWD fit-for-use assessment service.
About the meeting
The Duke-Margolis Institute for Health Policy’s annual State of Real-World Evidence Policy convening was held virtually on July 27, 2026. View the Duke-Margolis RWE Collaborative.