Optimizing Screening for Intrastriatal Interventions in Huntington's Disease Using Predictive Models.

Area of Science:

• Neuroscience
• Medical Imaging
• Machine Learning

Background:

• Intrastriatal delivery of therapeutics for Huntington's disease (HD) necessitates adequate caudate and putamen volumes.
• Current volumetric MRI is not standard in clinical practice and lacks data in large HD cohorts.

Purpose of the Study:

• To develop and validate predictive models for classifying HD patients exceeding striatal volume thresholds for intrastriatal therapy.
• To assess the feasibility of using machine learning models for prescreening in HD clinical trials.

Main Methods:

• Merged data from 1374 individuals across three HD cohorts (IMAGE-HD, PREDICT-HD, TRACK-HD/TRACK-ON).
• Utilized BORUTA algorithm to identify 10 key predictive variables from clinical data.
• Developed and compared random forest and logistic regression models to predict striatal volumes.

Main Results:

• The random forest model achieved an 83% area under the curve (AUC).
• The logistic regression model, using age, CAG repeat size, and symbol digit modalities test, yielded an 85.1% AUC.
• A probability cutoff of 0.8 resulted in low false positive (5.4%) and high false negative (66.7%) rates.

Conclusions:

• Machine learning models (random forest, logistic regression) effectively identify individuals with sufficient striatal volumes using accessible clinical data.
• These models can streamline clinical trial enrollment for HD and other neurodegenerative diseases with volume-based inclusion criteria.
• Implementation in prescreening can accelerate research and therapeutic development for conditions requiring volumetric assessments.