Understanding Further the Phenotypic Spectrum of Central Nervous System Inflammatory Demyelinating Disorders Using Unsupervised Clustering.

BACKGROUND

Central nervous system (CNS) inflammatory demyelinating syndromes, including multiple sclerosis (MS), aquaporin-4 antibody-positive neuromyelitis optica spectrum disorder (AQP4 + NMOSD), and myelin oligodendrocyte glycoprotein (MOG) antibody-associated disease (MOGAD), occasionally overlap. Some patients remain double-seronegative, showing atypical features that challenge current classifications.

OBJECTIVE

To better characterize the phenotypic spectrum of antibody-negative atypical inflammatory demyelinating disorders (AIDD) using unsupervised clustering.

METHODS

We retrospectively analyzed 316 patients (MS = 164, AQP4 + NMOSD = 36, double-seronegative NMOSD = 21, MOGAD = 15, AIDD = 80) followed between 2010 and 2023. Principal component analysis and k-means clustering were applied to AIDD cases using clinical, demographic, and radiological data.

RESULTS

AIDD patients had lower disability and fewer corpus callosum and posterior fossa lesions than MS and NMOSD. Three clusters emerged: (1) myelitis-predominant with unmatched CSF oligoclonal bands and longitudinally extensive spinal lesions, (2) brainstem-dominant with recurrent brainstem attacks, and (3) optic neuritis-dominant with recurrent LEON meeting MS dissemination criteria. Treatment patterns differed; rituximab was most frequent.

CONCLUSION

Double-seronegative AIDD represents a heterogeneous clinical spectrum. Unsupervised clustering provides a data-driven framework for refining phenotypic classification and may support biomarker and therapeutic development in antibody-negative CNS demyelination.