ML predicts MS outcomes

Multiple sclerosis can flare suddenly or worsen slowly over years, and a new pair of machine-learning tools aims to estimate that five-year risk for individual patients. (link.springer.com) The study, published April 19 in the *Journal of Neurology*, analyzed 34,510 adults with relapsing-remitting multiple sclerosis at baseline. The researchers built two models, called DAAE-M and ELIE, to predict who would worsen over the next five years using clinical data and treatment history. (link.springer.com) The team measured worsening in two ways: a clinical shift from relapsing-remitting disease to progressive disease, and a stricter measure called confirmed progression independent of relapse activity, or disability that accumulates without a recent attack. Over five years, 9.8% met the clinical progression definition and 21% met the objective progression definition. (link.springer.com; nationalmssociety.org) DAAE-M grouped patients into rising risk bands, with five-year clinical progression rates from 3.1% in the lowest band to 33.0% in the highest. On the objective measure, the same model ranged from 8.4% to 38.8% across risk groups. (link.springer.com) ELIE produced finer-grained deciles instead of broad bands, and five-year clinical progression climbed from 0.3% in the lowest decile to 21.5% in the highest. Objective progression rose from 0.9% to 32.5% across those deciles. (link.springer.com) The authors designed the two systems for different jobs. DAAE-M was built to be easier to inspect and use across software settings, while ELIE was built to update risk over time and better account for changing treatment histories. (pmc.ncbi.nlm.nih.gov) The treatment question sits at the center of the paper. In DAAE-M, high-efficacy disease-modifying therapy was associated with about half the progression risk of low-efficacy therapy, with reported risk ratios ranging from 0.42 to 0.59. (link.springer.com) That matters in a disease that still resists clean forecasting. About 85% of people with multiple sclerosis are initially diagnosed with relapsing-remitting disease, and many later move into a secondary progressive phase marked by accumulating disability. (nationalmssociety.org; nationalmssociety.org) The broader research field is also shifting away from rigid labels toward continuous measures of disease activity. A 2025 *Nature Medicine* paper used probabilistic machine learning on more than 8,000 trial participants and described multiple sclerosis as a severity spectrum rather than a set of fixed boxes. (nature.com) The new study does not settle how these scores should be used in clinic, and the authors frame them as prediction tools rather than treatment rules. But with 2.9 million people living with multiple sclerosis worldwide, a five-year forecast that holds up in practice could shape follow-up schedules, counseling and trial enrollment long before disability becomes obvious. (atlasofms.org; link.springer.com)