F013 Prognostic enrichment for early-stage Huntington’s disease using machine learning

<h3>Background</h3> The introduction of Huntington’s Disease Integrated Staging System (HD-ISS) facilitated highly targeted enrolment in HD clinical trials. However, each of the four stages in HD-ISS could benefit from prognostic enrichment, which aims to enhance the assessment of therapeutic effects by identifying high-risk individuals for inclusion. <h3>Aims</h3> We aimed to classify HD individuals into homogeneous groups based on forthcoming lateral ventricular enlargement rate (LVER). <h3>Methods</h3> We used data from 451 HD gene expansion carriers from previously published cohorts (PREDICT, TRACK, TrackON, and IMAGE). We applied whole-brain parcellation to longitudinal brain scans and used linear regression to measure LVER over 3 years (mm3/year). Subsequently, it was used as the target variable for our random forest stratification model. HD individuals were classified into two homogenous groups: low-risk and high-risk. The model was trained using different combinations of features at baseline including the genetic load, cognitive and motor biomarkers, as well as brain imaging-derived features. Proposed as a metric for prognostic enrichment, the probability of LVER surpassing a threshold of 750 mm3/year was compared with normalized prognostic index (PIN) figure 1. <h3>Results</h3> The stratification model had a cross-validated accuracy of 81%. Compared to the PIN score, the proposed probabilistic metric had lower Pearson’s kurtosis for HD-ISS stages 1 and 2 indicating higher discriminative power among individuals within these two stages. <h3>Conclusions</h3> This study validated ML-based predictive modelling for identifying high-risk HD individuals. The proposed method has the potential to enhance clinical trials by enabling more effective participant enrolment and improved post-hoc evaluation of therapeutic effects.