MRI-derived Subtypes Identified by SuStaIn Capture Distinct Clinical, Cognitive, and Disability Profiles in Multiple Sclerosis
Paolo Preziosa1, Loredana Storelli4, Elisabetta Pagani4, Nicolò Tedone2, Monica Margoni5, Federica Esposito6, Massimo Filippi3, Maria Rocca1
1Neuroimaging Research Unit, Division of Neuroscience, and Neurology Unit, 2Neuroimaging Research Unit, Division of Neuroscience, 3Neuroimaging Research Unit, Division of Neuroscience, Neurology Unit, Neurorehabilitation Unit, and Neurophysiology Service, IRCCS San Raffaele Scientific Institute; and Vita-Salute San Raffaele University, 4Neuroimaging Research Unit, Division of Neuroscience, 5Neuroimaging Research Unit, Division of Neuroscience, Neurology Unit, and Neurorehabilitation Unit, 6Neurology Unit, and Laboratory of Human Genetics of Neurological Disorders, IRCCS San Raffaele Scientific Institute
Objective:
To apply Subtype and Stage Inference (SuStaIn) algorithm to identify MRI-based subtypes that better stratify the heterogeneous clinical profiles in multiple sclerosis (MS).
Background:
Clinical MS classifications incompletely capture disease heterogeneity. Data-driven subtyping may uncover endophenotypes that better reflect MS pathobiology and explain variability in disability and cognition.
Design/Methods:
We retrospectively studied 1017 MS patients and 548 healthy controls who underwent standardized clinical, neuropsychological, and brain 3T MRI assessments. Z-scores of 21 MRI features, including white matter (WM) lesion volume, microstructural abnormalities of the major WM tracts, and normalized global and regional brain volumes, were input into SuStaIn to derive subtypes and stages. Associations with clinical phenotype (relapsing-remitting [RR] vs progressive), age at onset (AAO) (pediatric [POMS], adult [AOMS], late [LOMS]), disability (Expanded Disability Status Scale [EDSS]), and cognition (Brief Repeatable Battery of Neuropsychological Tests) were explored.
Results:
SuStaIn identified four MRI subtypes: lesion-led (44%), cortex-led (23%), tract-led (23%), and deepGM-led (10%). Subtypes differed by clinical phenotype and AAO (p<0.001). Cortex-led and deepGM-led were enriched in progressive forms (47% and 56%), whereas lesion-led and tract-led were mainly RRMS (66–71%). By AAO, tract-led was overrepresented in POMS (70/184, 38%) and underrepresented in AOMS (152/783, 19%), lesion-led predominated in AOMS (372/783, 48%), whereas deepGM-led was more frequent in LOMS (7/50, 14%). Disability milestones were reached more often in cortex-led and deepGM-led patients (EDSS≥4.0: 45% and 56%; EDSS≥6.0: 28% and 29%) than in lesion-led (34% and 20%) or tract-led (32% and 19%) (p<0.001). Although cognitive performance did not differ across subtypes (p≥0.513), within each subtype advancing SuStaIn stage was consistently associated with worsening global and domain-specific scores (p≤0.030).
Conclusions:
MRI-based subtypes from SuStaIn capture clinically meaningful heterogeneity in phenotype, age at onset, disability, and cognition. This data-driven approach may support personalized therapeutic strategies tailored to the pathological mechanisms underlying MS progression.
Disclaimer: Abstracts were not reviewed by Neurology® and do not reflect the views of Neurology® editors or staff.