A multimodal approach to distinguish multiple sclerosis phenotypes at diagnosis using biomarker profiles.

Abstract

Background: Multiple sclerosis (MS) is a complex and heterogeneous disease characterized by variable clinical outcomes.

Objective: We aimed to develop a predictive model combining principal component analysis (PCA) and clustering techniques to identify biomarker sets associated with MS and characterize distinct phenotypes.

Design: A monocentric, cross-sectional study on treatment naïve patients at the time of MS diagnosis.

Methods: Clinical, laboratory, and neuroimaging data were collected, including retinal layer measurements via optical coherence tomography and neurofilament light (NFL) chains levels.

Results: The cohort included 71 MS patients with mean age 35.7 years (SD = 9.8). PCA yielded five components with eigenvalues >1.0, explaining 68.1% of total variance. Component 1 showed strong negative coefficients for retinal thickness (ganglion cell-inner plexiform layer: -0.82, peripapillary retinal nerve fiber layer (RNFL): -0.79, macular RNFL: -0.75) and moderate positive coefficient for serum NFL (0.45). Component 2 featured high positive coefficients for NFL in cerebrospinal fluid (0.88) and serum (0.56). K-means clustering identified two distinct groups: one (n = 33) with thicker retinal layers, better cognitive performance, and unexpectedly higher serum NFL levels compared to the other group (n = 38).

Conclusion: These findings suggest that MS may present with distinct phenotypic profiles even at diagnosis. Future longitudinal studies are needed to validate these early biomarkers and refine personalized treatment approaches.