xHD-Vox, an Automated Speech Model for Estimating Motor and Cognitive Scores in Huntington Disease: Development and Longitudinal Validation.

Background: Huntington disease (HD) is a rare genetic neurodegenerative disease that causes progressive motor, cognitive, and psychiatric symptoms over decades after onset. Clinical care is typically provided in specialized centers with only annual clinical assessments, highlighting the need for more frequent and cost-effective monitoring. Objective: This study aimed to develop and validate xHD-Vox, a fully automated, interpretable, speech-based model for predicting the composite Unified Huntington Disease Rating Scale (cUHDRS) and its cognitive, motor, and functional components. Methods: We included 181 HD gene carriers (341 annual visits) from three French prospective cohorts: BIO-HD (NCT01412125), REPAIR-HD (NCT03119246), and MIG-HD (NCT00190450). Participants had ≥40 cytosine-adenine-guanine (CAG) repeats, available cUHDRS scores, and audio recordings of forward and backward counting (1-20). For model development and feature selection, we used a speech pathologist-annotated subset (145 visits and 90 participants). Selected speech features were then automated using Whisper, an open-source speech recognition tool. The final linear regression model, xHD-Vox, was calibrated on the training set of 269 visits (157 participants, and annotated subset included). Performance was evaluated on an independent longitudinal test set (24 participants, with 3 annual visits each) using mean absolute error, explained variance ( Results: Feature selection identified four key predictors: standardized CAG-age-product score, CAG repeat length, rate of numbers pronounced per second, and the SD of that rate. On the test set, xHD-Vox achieved a mean absolute error of 2.1 for cUHDRS and explained 57% of its variance, compared with 38% when using only demographic features. Longitudinal analyses using repeated-measures ANOVAs with post hoc Tukey tests confirmed a significant decline over the 2-year follow-up for both clinician-assessed measures and xHD-Vox predictions. At the group level, the mean 1-year and 2-year changes predicted by xHD-Vox were consistent with clinically measured changes, falling within the corresponding 95% CIs. Conclusions: We developed xHD-Vox, an interpretable and automated model that predicts clinical scores in HD using a short speech task. Predicted scores were consistent with clinician-assessed scores, supporting its potential use in mobile apps for remote monitoring. This approach could facilitate scalable, real-time tracking of disease progression, especially in underserved regions, and enable personalized and responsive clinical care.