Objective: To explore whether internet-based assessments combined with machine learning approaches can support scalable characterization of ataxia severity.

Background: In-person assessments of ataxia severity are limited by accessibility, scalability, and geographical diversity. These challenges may restrict participation and generalizability in both research and clinical contexts.

Method: We evaluated the feasibility, diagnostic precision, and generalizability of SARA-LeS, an abridged, self-administered online assessment of ataxia severity. SARA-LeS is a four-item questionnaire derived from SARA-Le, a previously validated, administrator-led videoconferencing assessment based on the eight-item Scale for the Assessment and Rating of Ataxia (SARA). Unlike SARA-Le, SARA-LeS is completed independently by patients via a simple web link on a personal device (phone or computer). SARA-Le was administered to 180 English- or Hebrew-speaking individuals with cerebellar ataxia across 70+ geographic locations in two countries: USA and Israel. This online data was compared with a large, expert-rated, in-person European cohort (EUROSCA; N = 710).

Results: The four-item SARA-LeS accounted for 96% of the variance in total SARA scores and demonstrated feasibility in 50+ participants. Bidirectional cross-dataset validation using machine-learning models showed high predictive performance withhin and between datasets (r = 0.91–0.99).

Conclusion: These findings suppot that SARA-LeS may provide an accessible and scalable approach for assessing ataxia across diverse populations.

« Back to 2026 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/accessible-assessment-of-ataxia-severity-across-diverse-populations-integrating-large-datasets-machine-learning-and-online-testing/