Using AI to Power a Real-World MS Disease Registry

Researchers from the Rocky Mountain Multiple Sclerosis Clinic (RMMSC) and Nira Medical recently presented a poster at the European Committee for Treatment and Research in Multiple Sclerosis about their efforts to take an AI approach to building a real-world MS disease registry to advance treatments and access to effective MS care.

Neurologist John Foley, M.D., founder of RMMSC, and Rebekah Foster, head of data at Nira Medical, recently responded via e-mail to Healthcare Innovation’s questions about their work. 

Could you briefly describe Nira Medical? Is it a partnership of independent neurology practices? Does it focus on the technology aspect?


Nira Medical is a national platform for independent neurology practices, built by and for clinicians. We’re not a tech company in the traditional sense, but technology is a major part of how we empower our physicians. Our platform supports independent neurologists with streamlined workflows, shared data infrastructure, and research capabilities that make it easier to deliver exceptional care.

The press release about this work quotes Rebekah as saying “Our goal was to harness the power of AI to overcome long-standing barriers in real-world MS research.” Can you talk about what some of those barriers are?

Historically, building MS registries has been a slow, manual process that depends on chart reviewers combing through unstructured clinician notes one by one. That limits scale, diversity, and the ability to capture the true complexity of this disease. The data we need like EDSS scores, relapse timelines, and MRI findings has always existed inside the EHR, but it’s been locked away in free text. 

Our biggest barriers were fragmentation, inconsistency, and time. Every clinic documents differently, which makes it difficult to aggregate and standardize data across thousands of patients. Using large language models, we can now extract and structure that data with accuracy and speed, turning what used to take months of manual work into hours. That unlocks a much larger, more representative dataset for understanding real-world outcomes in MS.

Could you talk about the way you would traditionally build an MS registry and how this approach differs from that? 

Traditionally, building an MS registry has meant pulling data manually from patient charts, one note, one lab result, one MRI at a time. Teams of abstractors would spend months reviewing records to extract things like EDSS scores or relapse events. It’s time-consuming, expensive, and inherently limited in scope, which means most registries end up covering a few hundred patients at best.


Using the Century Health Abstraction and Retrieval Model (CHARM) allows us to automate the extraction of structured variables from unstructured clinical records using Large Language Model (LLM) reasoning and clinical Natural Language Processing (NLP). With this approach, we can now process thousands of patient records in a fraction of the time.

The LLM extracts key variables with very high accuracy, and structures that data automatically. That means we can build a registry that’s not only larger and more diverse, but also continuously updated, a true living dataset that reflects what’s happening in real-world MS care.

Your work notes that AI has the potential to unlock the full potential of these insights by structuring and abstracting data at scale. Could you explain how that might unlock insights for researchers? 

When you can structure unstructured data at scale, you suddenly have access to thousands of real patient journeys instead of a handful. That means researchers can start asking and answering questions that were impossible before: How do certain therapies perform across subtypes? What early signs predict relapse? Which patients benefit most from B-cell therapies?

By turning messy EHR notes into analyzable data, AI allows us to surface these patterns quickly and continuously, rather than waiting years for manual chart reviews or small, curated studies. It gives researchers a living, real-world view of MS care.

How does Century Health come into this work? What is their CHARM model and what does it enable?
 
Century Health’s CHARM model is an AI-powered model that automates the extraction of structured variables from unstructured clinical records. The use of AI and the model’s high accuracy, scalability, and efficiency allow us to gather real-world evidence much faster than using manual abstraction, which remains slow, costly, and inconsistent.

CHARM automated these routine abstraction tasks and gives our partners reliable data sets at an unprecedented speed and scale. 

Does this have implications not just for drug development but also for other aspects of patient care?

While the initial application is in drug development and real-world evidence generation, the broader impact is on patient care itself. Every time we generate evidence faster and at a larger scale, we give clinicians clearer feedback loops about what’s working and for whom.
By structuring information that was previously buried in free-text notes or MRI reports, we can surface insights that show how patients are actually doing between visits: how disability evolves, how often relapses occur, or how individuals respond to therapy over time. Over time, this data infrastructure transforms evidence generation into a continuous learning system that directly improves care.

Do you think AI approaches like this could reshape how clinical researchers across disease states interact with real-world data and registries? 


What we’re seeing here is the beginning of a fundamental shift in how clinical researchers engage with real-world data. Historically, registries have been limited by what could be manually abstracted or entered into structured fields, so they were often small, retrospective, and time-intensive.

AI changes that equation. By automating the extraction of rich, disease-specific insights from unstructured EHR notes, models like CHARM make it possible to build and update registries in near real time, at a scale and depth that simply wasn’t feasible before.
This means researchers across therapeutic areas can move faster, from understanding treatment patterns and outcomes to identifying trial candidates and studying long-term effectiveness. Ultimately, it democratizes access to high-quality real-world evidence, bridging the gap between everyday clinical practice and research.

What is the next step in the development of this registry? 


The next step is expansion, in scope and scale. We’re working to extend this AI-powered registry framework beyond multiple sclerosis into other complex neurological conditions, such as myasthenia gravis, NMOSD, and Alzheimer’s disease, where real-world data are equally fragmented.

At the same time, we’re growing the size and depth of the MS registry itself by onboarding additional clinics and incorporating longitudinal data sources like imaging, labs, and patient-reported outcomes. Our goal is to build a living, learning registry that continuously evolves, capturing richer, more complete patient journeys and enabling both research and clinical insights that directly inform care.