Multiple Sclerosis: AI analysis prompts international reassessment of disease progression

Multiple sclerosis (MS) has long been regarded as a disease with different subtypes such as “relapsing” or “progressive.” An international study, published on August 20, 2025, in Nature Medicine under the leadership of the Medical Center – University of Freiburg and the University of Oxford, challenges this dogmatic model after analyzing the NO.MS cohort (study data from Novartis). Instead of fixed disease phenotypes, an AI-based model identifies four central state dimensions that better capture the progression of MS: physical disability, brain damage, clinical relapses, and silent inflammatory activity. These insights could fundamentally change the diagnosis and treatment of MS patients and may also be relevant for other diseases.

“Our data clearly show that MS cannot be characterized by subtypes such as relapsing or progressive MS, but is instead a continuous disease process with definable state transitions,” says Prof. Dr. Heinz Wiendl, Medical Director of the Department of Neurology and Neurophysiology at the Medical Center – University of Freiburg.

The findings are based on the analysis of more than 8,000 patients and over 35,000 MRI scans from various studies (NO.MS cohort, Roche Ocrelizumab cohort, MS PATHS cohort).

Disease as a dynamic system: A new view of MS

The probabilistic model describes MS as a sequence of states with specific transition probabilities. Early, milder states typically progress through inflammatory intermediate phases into advanced, irreversible stages of disease. Remarkably, direct progression into severe stages without prior inflammatory activity is virtually excluded—silent, symptom-free inflammation or relapses are the central drivers of deterioration.

Implications for diagnosis, therapy, and approvals

The previous classification system often hinders access to effective medications, as approvals are based on rigid subtype definitions. The new model enables individualized risk assessment—independent of the diagnosed subtype.

“Instead of categorizing patients, we should quantify their state and track it dynamically,” Wiendl emphasizes. Patients with active but clinically silent inflammatory activity, in particular, require early treatment decisions, as the model strikingly demonstrates.

A broadly applicable model—beyond MS

State-based modeling using artificial intelligence methods is not only a scientific breakthrough in MS research. “The principle is fundamental and pioneering—and it can also be applied to many other diseases, both within neurology and beyond,” says Prof. Dr. Lutz Hein, Dean of the Faculty of Medicine – University of Freiburg. The key is to move away from rigid, predefined disease categories and instead focus on data-driven, flexible disease states within an illness.

Next steps: Translation into clinical practice and research

“The next important step is to transfer these possibilities of individualized risk assessment into clinical practice and to collect prospective data,” emphasizes Prof. Dr. Peter Berlit, Secretary General of the German Society of Neurology. The model has already been successfully validated within the study using external clinical and real-world datasets. The next phase is integration into everyday clinical practice—for example, in treatment decisions or to improve patient education. In the long term, dynamic classification could also fundamentally change the regulatory logic for future therapies.