Anatomical categorization of isolated non-focal dystonia: novel and existing patterns using a data-driven approach.

Abstract

According to expert consensus, dystonia can be classified as focal, segmental, multifocal, and generalized, based on the affected body distribution. To provide an empirical and data-driven approach to categorizing these distributions, we used a data-driven clustering approach to compare frequency and co-occurrence rates of non-focal dystonia in pre-defined body regions using the Dystonia Coalition (DC) dataset. We analyzed 1,618 participants with isolated non-focal dystonia from the DC database. The analytic approach included construction of frequency tables, variable-wise analysis using hierarchical clustering and independent component analysis (ICA), and case-wise consensus hierarchical clustering to describe associations and clusters for dystonia affecting any combination of eighteen pre-defined body regions. Variable-wise hierarchical clustering demonstrated closest relationships between bilateral upper legs (distance = 0.40), upper and lower face (distance = 0.45), bilateral hands (distance = 0.53), and bilateral feet (distance = 0.53). ICA demonstrated clear grouping for the a) bilateral hands, b) neck, and c) upper and lower face. Case-wise consensus hierarchical clustering at k = 9 identified 3 major clusters. Major clusters consisted primarily of a) cervical dystonia with nearby regions, b) bilateral hand dystonia, and c) cranial dystonia. Our data-driven approach in a large dataset of isolated non-focal dystonia reinforces common segmental patterns in cranial and cervical regions. We observed unexpectedly strong associations between bilateral upper or lower limbs, which suggests that symmetric multifocal patterns may represent a previously underrecognized dystonia subtype.