Key Takeaways
- AI is helping MedTech companies turn messy real-world data from electronic health records, claims, and lab records into usable real-world evidence for product development and regulatory decisions
- Johnson & Johnson, working with Mayo Clinic, Mercy Health, and Yale, secured the first FDA label expansion for a medical device based solely on a comparative real-world evidence study using EHR databases
- The FDA has authorized over 1,600 AI-enabled medical devices for marketing, a count that raises questions about how regulators will keep evaluation standards current
- Natural language processing lets researchers pull specific details, like wound size or healing rate, out of unstructured health records instead of relying on generic diagnosis codes alone
- A meta-analysis of 25 peer-reviewed studies published in the Journal of Robotic Surgery, reviewing studies conducted between 2024 and 2025, found AI-assisted robotic surgery reduced operative time by 25% and intraoperative complications by 30% compared to manual methods
Medical technology has always generated more than just devices. Every heart pump, catheter, and surgical tool produces a trail of information about how it performs once it leaves the lab and enters real clinical settings. That trail, known as real-world data, is now meeting artificial intelligence in ways that are changing how MedTech products get built, tested, and approved.
Over 1,600 AI Devices FDA-Authorized
The FDA has authorized over 1,600 AI-enabled medical devices for marketing in the United States, supporting disease detection, diagnosis, and treatment planning. That count shows how deeply AI has already embedded itself into the tools clinicians use every day.
What’s less visible, but arguably more important, is how AI is reshaping the evidence behind those devices. Real-world data, pulled from electronic health records, insurance claims, and laboratory systems, becomes real-world evidence once it’s analyzed and interpreted.
MEDDDICAL, a health market intelligence company that works at this intersection, says that the stakes are high because real-world evidence touches nearly every part of the product lifecycle: manufacturing decisions, performance monitoring, safety signal detection, and the regulatory approval process itself. Getting it right can speed up timelines and expand what a device is approved to do. Getting it wrong can mean missed safety issues or evidence that regulators won’t accept.
Why Real-World Data Is So Messy
Raw healthcare data rarely arrives ready to use. Electronic health records, claims databases, and lab systems were built for billing and clinical documentation rather than research, so the information inside them is often unstructured, incomplete, or inconsistent from one hospital system to the next.
From Raw Data to Usable Evidence
Turning that raw material into trustworthy evidence takes real work. Device identifiers are frequently missing or inconsistent across systems, diagnosis and procedure codes don’t always capture what actually happened to a patient, and long-term outcomes can fall out of the record entirely once a patient moves between health systems. A device might show up in the data without ever being named or clearly identified, leaving researchers to guess what was actually used in a procedure.
Bias presents another quiet threat. Data can carry ethical, gender-based, or commercial bias baked in from the start, and a dataset that looks complete on the surface can still skew who gets represented and who doesn’t. Sorting through that requires people who know what to look for and where the fault lines typically run.
How AI Fills the Gaps
This is where artificial intelligence earns its keep. Natural language processing can scan unstructured clinical notes and correlate scattered references with specific devices, reconstructing information that would otherwise be lost in free text. Machine learning models go a step further, analyzing large patient datasets to flag patterns and predict outcomes that would take human reviewers far longer to find on their own.
Large language models are also opening this kind of analysis to a broader set of professionals. Clinicians, engineers, and product teams who never touched a line of code can now query datasets using plain language, while newer agent-based analytics tools reduce the need for manual coding. As medical datasets grow through record linkage and device-generated data streams, accessible, AI-powered analysis becomes less of a convenience and more of a necessity.
Winning Label Expansions with RWE
Real-world evidence has become a genuine path to expanding what a device is approved to treat, especially when running a traditional clinical trial isn’t practical.
A First-of-Its-Kind EHR-Based Approval
Surgical devices face a particular challenge: enrolling large numbers of surgery patients in randomized trials is difficult, and the patients who do enroll may not represent the broader population who’ll eventually use the device. Johnson & Johnson, working alongside Mayo Clinic, Mercy Health, and Yale, found a way around that bottleneck. Through a test case funded by the FDA’s National Evaluation System for health Technology, the partnership became the first MedTech company to receive an FDA-approved label expansion based solely on a comparative real-world evidence study built from EHR databases, applied to a device for patients with persistent atrial fibrillation.
That success didn’t stay isolated. Lessons from the atrial fibrillation study informed a second comparative study, which led to a zero-fluoroscopy workflow label expansion covering eleven separate devices, allowing ultrasound-based imaging and navigation to serve as an alternative to X-ray guidance during procedures.
Iterative Device Upgrades Need Less Trial Data
Not every device needs to prove itself from scratch. Many MedTech approvals represent iterative improvements on tools that have been safely used for generations. Scalpels are a good example: people have relied on them for roughly five thousand years, so a new version doesn’t necessarily need a randomized controlled trial. Instead, it can be tested against prior versions in preclinical settings, since the underlying safety profile is already well understood. Real-world evidence fits naturally into this pattern, giving manufacturers and regulators a way to validate incremental changes without re-running the kind of large trial a brand-new device would require.
Weighing Benefits Against Risks
Benefit-risk assessment sits at the center of every device approval, requiring regulators and manufacturers to weigh a device’s positive effects against its potential downsides. Pharmaceuticals typically generate benefit-risk data through a Phase 3 trial, but medical devices don’t face the same trial requirements, so real-world evidence often has to fill that gap directly, identifying rates of both benefit and risk from data gathered after a device is already in use.
Bringing Patient Preference into the Equation
Patient perspective adds another layer to that calculation. Understanding how patients themselves weigh a device’s benefits against its risks helps manufacturers evaluate whether a technology is worth the trade-offs it asks patients to accept. The FDA has updated its guidance on incorporating this kind of patient preference information across a device’s full product lifecycle, reinforcing that patient voice belongs in the evidence base alongside clinical data. Surgeons benefit from this evidence too, since it reveals how a given device performs across different clinical settings and patient types well before a particular case begins.
AI Reshaping the Operating Room
Surgery itself has become a major source of real-world data, not simply a place where devices get used.
Surgical Data Rivers Fuel Predictive Models
Modern surgical systems now measure details that once went unrecorded entirely: how much force was applied, how much power was used, and the precise angles involved in a given maneuver. Johnson & Johnson is among the companies combining AI, large datasets, and analytics to make procedures safer and more efficient, treating this continuous stream of surgical information, sometimes described as a data river, as fuel for increasingly sophisticated predictive models. NVIDIA has pursued a similar direction, applying AI to real-time analysis during surgery itself, extending into surgical decision-making, medical education, and coordination among operating room teams.
Measurable Gains in Time and Complications
The payoff from this shift is becoming measurable rather than theoretical. A meta-analysis of 25 peer-reviewed studies published in the Journal of Robotic Surgery, covering research conducted between 2024 and 2025, found that AI-assisted robotic surgeries delivered a 25% reduction in operative time and a 30% decrease in intraoperative complications compared to manual methods, alongside notable gains in surgical precision. Those are the kinds of numbers that get MedTech leaders’ attention, since they translate directly into shorter procedures, fewer complications, and better use of operating room time.
Regulators Race to Keep Pace
Regulatory bodies are not standing still while this technology matures. Agencies around the world are actively updating how they evaluate AI-generated evidence, recognizing that older frameworks weren’t built for this volume or speed of data.
FDA Evaluates RWD Quality for RWE
The FDA has issued final guidance on the use of real-world evidence to support regulatory decision-making for medical devices, building on earlier draft guidance. The agency has also cited specific examples of AI and machine learning-based devices validated using real-world data, including systems assessed against intensive care unit records and electrocardiogram data, and has sought public comment on measuring and evaluating AI-enabled medical device performance in the real world, including strategies for identifying and managing performance drift.
Human Oversight Still Required
Regulators outside the United States are moving in a similar direction. Canada’s Drug Agency (CDA-AMC) has issued new guidance on reporting AI methods used to generate real-world evidence for health technology assessments, with further updates and bulletins following. The message is consistent: AI should supplement established methods rather than replace them, and human oversight remains a requirement rather than an option. That stance reflects a broader theme running through regulatory thinking in 2026 – AI can accelerate evidence generation dramatically, but it still needs informed people checking its outputs for bias, drift, and plain old mistakes.
AI and RWE Are Redefining MedTech Development
Across label expansions, surgical innovation, and benefit-risk assessment, the pattern is consistent: AI gives MedTech teams the ability to turn scattered, imperfect real-world data into evidence that regulators, clinicians, and patients can trust. That shift raises the bar for both data governance and human judgment, since faster analysis only helps when the underlying data quality holds up.
For MedTech leaders and pharmaceutical companies working through this environment, the practical move is building evidence strategies around real-world evidence that advisories like MEDDDICAL can analyze. This can help keep pace with where regulators and AI capabilities are both heading next.
MEDDDICAL
Aptos 221
Edificio D2C
Sotogrande
Cadiz
11310
Spain
