David Talby, PhD, MBA, CEO at John Snow Labs. Solving real-world problems in healthcare, life sciences and related fields with AI.gettyThe legal and regulatory commentary on the FDA's December 2025 final guidance on real-world evidence (RWE) for medical devices has focused on what the guidance asks sponsors to prove. Less discussed are two structural changes the guidance makes that genuinely widen the universe of what can be submitted to the agency in the first place. Together, they are incredibly important elements of the new guidance, and the reason the demanding evidence requirements that come with it are worth meeting.​The First Change: De-Identified Data Is Now In ScopeThe 2017 RWE guidance, which the new document supersedes, generally expected sponsors to be able to provide identifiable, individual participant-level data on request. That requirement made large de-identified databases practically unusable for regulatory submissions. Most multi-institutional registries, most claims databases at scale and most cross-institutional research datasets were often difficult or impractical to use in regulatory submissions at scale. The 2025 guidance changes that. For certain device submissions, the FDA will accept real-world evidence without requiring identifiable patient data in the marketing submission.That single change opens up an enormous pool of data that was previously off limits to regulators. National disease registries, payer claims at scale, federated research networks and international real-world data sources all become candidates. The catch is that the guidance is not lowering the quality bar. De-identification is no longer the price of admission to compliant data use generally; it is the price of admission to the larger evidence pool the FDA will accept, and it has to be done in a way the agency can rely on.​The Second Change: Real-World Evidence Can Stand On Its OwnThe other shift in the 2025 guidance is one that regulatory analysts have been waiting on for years. Historically, real-world data has been treated as supporting context: useful for hypothesis generation, post-market surveillance or supplementing a clinical trial, but rarely accepted as primary evidence for a regulatory decision. The new guidance changes that posture. RWE may, in appropriate circumstances, contribute substantial or even primary evidence supporting a regulatory decision​.​That is a significant change in what a regulatory submission can look like. For certain device questions, a sponsor can build a submission around a well-designed real-world analysis rather than around a randomized trial supplemented by real-world data. The economics change, as do the timelines, the scope of feasible studies and the kinds of regulatory questions that can be answered.​What The Two Changes Do TogetherThe two shifts compound. Larger, more representative, de-identified datasets become eligible, and those datasets can carry the primary regulatory argument. For diseases where randomized trials are slow, expensive or ethically constrained, that combination is genuinely new. External control arms, comparative effectiveness studies against real-world standard of care, post-market expansions of indication and broader safety analyses all become more tractable.The change is not just a regulatory adjustment. It changes which research questions can be answered by which methods, and it does so at a moment when the engineering tools to produce regulatory-grade real-world evidence (including extraction from clinical narratives, multimodal integration, fact-level provenance and reproducibility at audit) have matured to the point where the underlying data can be made fit for purpose.​The Bar That Comes With The DoorNone of this lowers what the FDA is asking for. If anything, it raises it. The wider the data pool the agency is willing to consider, the more carefully it has to evaluate the relevance and reliability of any particular dataset. In practice, that means two things:1. The clinical facts in a submission have to accurately represent what happened to the patient. Structured EHR fields and claims data alone cannot deliver that.2. Every fact has to be traceable to its source document, scored for confidence and reproducible from versioned models and rules years later. Both bars apply with full force to de-identified datasets serving as primary evidence.That has a specific implication for de-identification itself. Under the old guidance, de-identification was a privacy hygiene concern. Under the new one, it is a regulatory pathway requirement. A submission that uses de-identified RWE as primary evidence has to demonstrate that the de-identification process is itself defensible: PHI removed at regulatory-grade accuracy, the clinical signal the analysis depends on preserved and the entire process auditable from the same source-to-fact lineage the rest of the platform supports.​That has practical implications for how de-identified real-world evidence is prepared for regulatory use. De-identification may need to account for multiple data modalities, including clinical text, scanned documents and imaging metadata, because protected health information can appear in more than structured fields alone. Sponsors may also prefer to perform processing within environments that preserve data governance controls and minimize unnecessary movement of identifiable information, particularly when working across institutions or jurisdictions.Because the FDA's focus remains on whether evidence is relevant, reliable and fit for purpose, sponsors using de-identified datasets should be prepared to document how de-identification was performed, how data utility was preserved and how the resulting evidence remains traceable and reproducible. Independent evaluation or validation may strengthen confidence in the process where appropriate, especially for higher-stakes submissions.Where de-identification is used as part of a regulatory strategy, approaches such as HIPAA Safe Harbor or Expert Determination should be applied with appropriate expertise and documentation so that sponsors can support the credibility and integrity of the resulting evidence package.​