Machine learning enables personalised prediction of cognitive decline in MS

An Amsterdam-based group identified 3 distinct cognitive phenotypes (CPs) in multiple sclerosis (MS), each with unique progression patterns, using machine learning (ML). These findings highlight the potential for personalised prediction of cognitive decline and its impact on disability in MS.

Cognitive impairment affects 40-70% of patients with MS, across all phenotypes. Patients may experience multiple cognitive deficits, which can impact processing speed, attention, executive function, working memory, and visuospatial memory. Dr Elisa Colato (Amsterdam UMC, the Netherlands) noted that recent studies have applied ML to identify cross-sectional CPs in MS, but did not capture how cognitive damage evolves. Dr Colato and colleagues aimed to identify data-driven CPs and their progression patterns in MS [1]. They used Subtype and Stage Inference (SuStaIn), an unsupervised ML approach combining clustering with disease progression modelling. By assessing associations with future disability, the clinical relevance of CPs was determined.

Using retrospective, longitudinal data from 3,407 MS patients, the model was validated internally in a training cohort and externally in an independent cohort. 3 cognitive phenotypes were identified, with a distinct progression pattern:

1. Verbal memory-led;
2. Verbal fluency and visuospatial/processing speed-led;
3. Executive function and processing speed-led.

The longitudinal stability of these phenotypes was assessed in both cohorts, with follow-ups of 2 years (training) and 1 year (external). In the training cohort (n=779), the initial subtype was maintained by 94% at 1 year, 57% at 3 years, and 37% at 5 years. In the external cohort (n=493), stability rates were 93%, 84%, and 51%, respectively. In the executive function and processing speed-led group, which also showed the highest rate of cognitive impairment, there was a significant association with future disability (P<0.05).

Dr Colato concluded that these findings may improve disease monitoring and inform the development of tailored interventions. Further research is needed to identify the underlying patterns of cognitive damage.

1. Colato E, et al. Identifying cognitive phenotypes and their progression patterns in multiple sclerosis using a machine learning model. O091, ECTRIMS 2025 Congress, 24-26 September 2025, Barcelona, Spain.

Medical writing support was provided by Michiel Tent.

Copyright ©2025 Medicom Medical Publishers



Posted on 29 September 2025