A machine learning-derived speech index as a biomarker for Huntington's disease severity

Area of Science:

• Neurology
• Biomarker Discovery
• Digital Health

Background:

• Huntington's disease (HD) requires objective biomarkers for treatment development.
• Current methods like clinical scales and neuroimaging have limitations in accessibility and cost.
• Automated acoustic analysis offers a potential solution for scalable HD assessment.

Purpose of the Study:

• To develop and validate a novel Speech Index for stratifying HD stages and predicting disease severity.
• To establish a sensitive, scalable, and objective digital biomarker for HD.
• To overcome limitations of current clinical and neuroimaging assessments.

Main Methods:

• Recruited 141 HTT gene carriers and 69 healthy controls.
• Extracted 28 speech features from standardized passage recordings.
• Constructed a composite Speech Index using Bootstrap LASSO Regression and validated against clinical and neuroimaging data.

Main Results:

• The Speech Index significantly increased with HD progression (p < 0.001).
• Strong correlations were observed between the Index and clinical measures (cUHDRS, TMS, SDMT) and neuroimaging (caudate, putamen volumes).
• Generalized additive models showed high predictive value (pseudo-R² 0.430–0.596).

Conclusions:

• A fully automated and interpretable Speech Index was developed as a valid digital biomarker for HD severity.
• The Speech Index shows promise for remote monitoring and clinical trial applications.
• This biomarker offers objective assessment of therapeutic efficacy in Huntington's disease.