by Agne Straukiene
For our Paper of the Month for October, we have chosen Ganjgahi, H., Häring, D.A., Aarden, P. et al. AI-driven reclassification of multiple sclerosis progression. Nat Med (2025). https://doi.org/10.1038/s41591-025-03901-6
Multiple sclerosis (MS) affects nearly 3 million people worldwide, yet the traditional categories of relapsing-remitting, secondary progressive, and primary progressive do not fully describe the biological variation or course of the disease. To address this, Ganjgahi et al. analysed the Novartis–Oxford MS database, which included over 8,000 patients, 118,000 visits, and 35,000 MRI scans. Clinical data comprised Expanded Disability Status Scale (EDSS) scores, walking speed (T25FWT), manual dexterity (9HPT), cognitive performance (PASAT), and reported relapses. MRI measures included T2 lesion volume, gadolinium-enhancing T1 lesions, and brain volume. Progression on MRI was defined by new or enlarging T2 lesions, new Gd+ lesions, and brain atrophy.
The study used a probabilistic factor analysis hidden Markov model (FAHMM). This statistical approach combined MRI and clinical measures to identify patterns of disease expression and to estimate the probability of transition between different disease states over time. The model was trained and then validated in independent datasets from clinical trials (including ocrelizumab studies) and from the real-world MS PATHS cohort involving more than 4,000 patients.
Eight disease states were identified and grouped into four broader categories. Early/mild/evolving states (1–3) included ambulatory patients with low disability and limited lesion burden. Asymptomatic radiological activity (state 4) reflected active MRI inflammation without clinical symptoms. Relapse (state 5) described acute clinical relapses, which in some cases were not accompanied by radiological progression. Advanced states (6–8) were defined by higher disability, cognitive decline, substantial atrophy, and a low probability of improvement. Patients generally moved from early/mild to advanced stages through inflammatory states, underlining the role of both symptomatic and silent inflammation in driving long-term disability.
The strengths of this work include the large sample size, integration of multimodal data, and validation across trial and real-world populations. The approach offers a more detailed description of how patients move between different phases of MS than the traditional subtype system. However, there are limitations: trial populations may not represent all patients seen in practice; only conventional MRI measures were used, without spinal cord or cortical imaging; and fluid biomarkers such as NfL or GFAP were not included. These factors mean further validation in broader clinical settings will be important.
In summary, the study describes MS as a continuum of disease states rather than fixed subtypes, with inflammation as a key driver of progression. This staging system may support more accurate monitoring of disease evolution and highlight areas where new treatments are most needed.
