Commentary|Videos|September 26, 2025

Leveraging Artificial Intelligence to Extract Real-World Insights in Multiple Sclerosis: Rebekah Foster, MBA & John Foley, MD, FAAN

At ECTRIMS 2025, a duo of experts discussed using AI-powered unstructured data processing to enhance understanding of drug efficacy, safety, and patient outcomes in multiple sclerosis. [WATCH TIME: 3 minutes]

WATCH TIME: 3 minutes | Captions are auto-generated and may contain errors.

"From a clinical point of view, we're already learning things that were inaccessible to us previously. One of the key takeaways is that we can definitely leverage artificial intelligence in a very productive way for clinical benefit. We'll see that in the next few years unfold in a fairly dramatic way."

Electronic health record (EHR) data for patients with multiple sclerosis (MS) are often difficult for clinicians to access because of fragmentation across systems and the predominance of unstructured documentation. Traditional real-world registries typically rely on manual chart review or structured fields that may omit key clinical information, such as imaging findings, lesions, and disease subtypes, limiting cohort size and representativeness. Artificial intelligence (AI) and machine learning (ML) tools have been proposed to structure and abstract these data at scale, enabling more comprehensive analyses of patient outcomes and treatment patterns.

In a new study presented at the 2025 European Committee for Treatment and Research in Multiple Sclerosis (ECTRIMS) Congress, held September 24-26, in Barcelona, Spain, an AI-driven approach demonstrated the feasibility of automating EHR curation to generate large-scale real-world evidence for MS. Researchers optimized the model, called CHARM (Century Health Abstraction and Retrieval Model), to extract structured data from unstructured clinical notes and MRI reports from a deidentified EHR database in Utah from 2011 to 2025. 

The cohort included 4179 patients with MS (women, 73.8%; mean age, 51.3 years), of whom 84.1% received at least 1 B-cell therapy. Variables abstracted in the model included MS diagnosis, disease subtype, therapy details and discontinuation reasons, estimated Expanded Disability Status Scale (EDSS), relapses, and MRI results. Presented by lead author John Foley, MD, FAAN, findings revealed that estimated EDSS was successfully derived for 97.9% of visits, with a median score of 3.0; 79.5% of patients with MS experienced at least 1 relapse.

At the Congress, Foley, founder and CEO at the Rocky Mountain MS Clinic, and coauthor Rebekah Foster, MBA, head of data at Nira Medical, spoke with NeurologyLive® To discuss the potential of AI to transform clinical research and patient care in MS. By applying unstructured data processing to patient charts, the duo noted that providers can now derive insights that were previously labor-intensive or inaccessible. During the conversation, Foley and Foster also emphasized the ability to generate synthetic EDSS scores for disability progression, expanding the scope of clinical insights beyond traditional trial settings.

Click here for more coverage of ECTRIMS 2025.

REFERENCES
1. Foley J, Goldenberg J, Foster R, Srivastava V, Hariharan S. A Novel Large Language Model Approach for Building a Real-World Multiple Sclerosis Disease Registry. Presented at ECTRIMS Congress; September 24-26, 2025; Barcelona, Spain. Abstract P365.

Newsletter

Keep your finger on the pulse of neurology—subscribe to NeurologyLive for expert interviews, new data, and breakthrough treatment updates.


Latest CME