Modeling Huntington’s disease progression utilizing the Huntington’s disease integrated staging system

Objectives: Huntington’s disease (HD) is a rare, progressive, neurodegenerative disorder that has faced a large number of failed trials despite promising therapeutics. One theory as to why potentially promising therapies have failed is that the therapeutic window is relatively early in disease progression, so trials may have failed due to inclusion of patients past the therapeutic window. The Huntington’s disease integrated staging system (HD-ISS)[ 1] offers a way to help enrich trials for earlier stage patients. To aid trial design and help understand how effectively disease trajectories in different HD-ISS stages can be predicted with a model, we leverage data from Critical Path Institute’s (C-Path’s) aggregated HD data to build a longitudinal disease progression model utilizing HD-ISS as a covariate. The model is planned to eventually be expanded into a clinical trial simulation platform. Methods: Models were explored using NONMEM, Pirana, and PsN and compared using diagnostic plots, visual predictive checks (VPCs), and 5-fold cross validation. Aggregated data from four clinical trials and four observational studies comprising over 16,500 HD patients from C-Path's database was used for modeling. Data was filtered to only include patients with at least 40 CAG repeats in the Huntingtin gene, the cutoff for complete penetrance in HD, and censored at 5-years post baseline. The data lacked the putamen and caudate information needed to distinguish HD-ISS stage 0 from stage 1, so these two stages were combined into a single category for modeling. Longitudinal total motor score (TMS) from the composite Unified Huntington’s Disease Rating Scale (cUHDRS)[ 2] was used as the dependent variable as it is a common endpoint in clinical trials. Four different model structures were considered: a generalized logistic model, a generalized logistic model with model-fit upper and lower bounds, a beta regression model with logit link, and a bounded integer model [ 3, 4]. VPCs were used to evaluate how well models predicted longitudinal TMS for the different HD-ISS stages at the upper 95%, lower 5%, and mid-ranges of the data. Results: The bounded integer model showed the best model fit based on 5-fold cross validation and VPCs stratified by HD-ISS stage. The VPCs indicated that the bounded integer model had good fit at all TMS ranges for all HD-ISS stages, with the caveat that stages 0 and 1 were evaluated together as a single category due to data limitations. Conclusions: TMS at all HD-ISS stages can be accurately predicted using a bounded integer model. This is promising for trial simulation and planning, and especially for sponsors that want to enrich on earlier HD-ISS stages. Effective prediction and trial planning at early HD-ISS stages built into a clinical trial simulation platform has potential to help encourage continued investment into therapeutics for HD.