Is MS Subtyping Clinically Meaningful? Experts Weigh In

WEST PALM BEACH — There is no shortage of efforts to subtype multiple sclerosis (MS) using data from imaging, molecular activity, genetics, epigenetics, and other factors. However, there is ongoing deliberation about the extent to which MS subtyping will advance clinical care.

In a recent debate at the Americas Committee for Treatment and Research in Multiple Sclerosis Forum 2025, both debaters agreed that traditional phenotypes, primarily based on patterns of relapse and progression, are entirely inadequate. 

Frustrated, Heinz Wiendl, MD, director of the Department of Neurology, University of Freiburg, Breisgau, Germany, spoke first, and he made that point clear.

The classically defined relapsing-remitting and progressive forms of MS “poorly align with MS pathology and have limited value for prognostication,” Wiendl said. Their limited clinical utility is the reason to pursue better subtyping of a disease heterogenous in almost every sense from disease presentation to clinical course. He believes that there has been substantial progress.

Definitive MS Subclasses Elusive

His opponent, Katherine Fitzgerald, ScD, associate professor of neurology at Johns Hopkins University, Baltimore, Maryland, acknowledged that MS subtypes are an appealing concept but argued that none of the existing subtyping approaches have successfully defined exclusive patient subgroups. So far, all identified subgroups have been relative rather than absolute. 

“We can define broad endophenotypes,” agreed Fitzgerald, but added that they overlap. While some patients might fit better into one endophenotype, they frequently have some features of another. The result is that a subtype cannot be assigned with 100% probability that any individual patient will not behave like another subtype, she added. 

Based on subgroups proposed to date, “a spectrum model is likely to be a better fit” if subclasses of MS are applied to prognosis, first- and second-line therapies, or other applications meaningful to patient care, Fitzgerald argued.

Fitzgerald’s argument focused on current and future potential of redefining MS as a group of related diseases with distinct processes and outcomes

Wiendl, a frequent contributor to research on MS subtyping, disagreed with this perspective.

“I am not even sure why we are having this debate. It is pretty obvious that subtypes exist,” said Wiendl, who believes many subtypes are — or will become — actionable. As an example, he pointed to MS patients unresponsive to interferon, a subgroup that “is opening the window to a better understanding of the underlying immune alterations,” while also highlighting the importance of individualized therapy. 

A Hindrance to Personalized Care?

Although he conceded that individualized treatment on the basis of immunobiology might not be ready for widespread application, he cited research published last year that supported the concept of endophenotyping with blood immune signatures. 

In this paper Wiendl and colleagues, described three distinct peripheral blood immunological endophenotypes that were unequally responsive to therapies, including interferon- b , when long-term follow-up data were evaluated.

The proof-of-concept study is likely a harbinger of more ambitious efforts based on big data. He cited several studies employing artificial intelligence to look for patterns across all variables in a large number of patients with the hope of reclassifying MS.

“One example is the NO.MS [Novartis-Oxford database] cohort, where probabilistic machine learning was applied to classify multiple sclerosis based on four dimensions, namely: brain damage, clinical relapses, sub-clinical disease activity, and physical disability,” Wiendl reported.

The clinical application of this information is still being explored, but by integrating data from multiple sources, he expects the likelihood of identifying meaningful subtypes to increase significantly. The excitement surrounding machine learning lies in its ability to analyze complex variables beyond the scope of traditional analytical methods. 

For her part, Fitzgerald reviewed a long list of potentially promising methods of subclassifying MS, including familial markers of disease, transcriptomic analyses, immunological biomarkers, and polygenic risk scores, and noted that none have fully discriminated MS patients for prognosis and outcome. 

Fitzgerald expressed concern about using endophenotypes in clinical practice when they are not mutually exclusive. Given the clinical heterogeneity of MS, broadly grouping patients based on shared characteristics could hinder the personalized care needed to address distinct disease features. 

“If we are defining distinct categories using continuous measures, we are probably losing some information,” she said. This extends to the application of categories created by marching learning from large datasets when offering care to “a real-world population.”

Wiendl reported financial relationships with Bayer, Biogen, CSL Behring, EMD Serono, Fresenius, Merck, Novartis, Omniamed, Roche, Sanofi, and Teva. Fitzgerald reports a financial relationship with SetPoint Medical.