Neurodegenerative disorders, including amyotrophic lateral sclerosis (ALS), Friedreich's ataxia (FA), multiple sclerosis (MS), Duchenne muscular dystrophy (DMD), Alzheimer?s disease (AD), Parkinson?s disease (PD), and Huntington?s disease (HD) are characterized by heterogeneous disease progression. Efforts to identify responder subgroups may uncover subgroups that are more homogeneous in disease-related features than the full study population. As a result, a subgroup may exhibit a statistically significant effect size. However, current methodologies for subgroup analysis are limited by the relatively small number of prognostic and predictive indicators that can be used to describe subgroups. These methods are not well suited to describing subgroups with reduced heterogeneity in disease progression, or in identifying indicators for multifactorial diseases. We have developed and submitted a patent application for a novel subgroup analysis method based on grouping participants with similar predicted disease progression profiles and analyzing nearest neighbor subgroups within a clinical trial. We call this method Detectable Effect Cluster(DEC) analysis. In our phase 1 and phase 2 SBIR grants, we used ALS as a model disease to develop our API product that uses machine learning disease models to improve trial arm randomization and provide covariates for statistical analysis. In the ongoing phase 2 grant we are expanding our disease offerings to include AD. PD and HD. Building on a set of ALS disease progression models that we have previously developed and validated, we seek in this grant application to develop a novel prototype machine-learning based subgroup analysis application that we plan on adding to our product offerings. During this proposed phase 1 grant, we will address research-level questions regarding the nature of the subgroups defined using DEC analysis including how to define confidence intervals of our DEC-clusters, and estimated bounds for using prediction-thresholds as selection criteria for a confirmatory clinical trial. Finally, we will apply DEC analysis to three publicly-available clinical trial data sets in an attempt to identify subgroups with significant treatment effects.
Aim 1 : We will apply methods used in image analysis for identifying statistically significant subgroups to address the multiplicity issue inherent in DEC Analysis.
Aim 2 : We will use statistical methods to model the confidence intervals of a power analysis in which DEC cluster based selection criteria would be used for a confirmatory trial.
Aim 3 : We will isolate records from PRO-ACT that include whether a patient was treated with riluzole and two other publicly available recent ALS datasets to test the application of DEC Analysis. Origent?s current suite of products will answer drug development needs of a full portfolio of neurodegenerative diseases. Ultimately, we see a series of machine learning applications aimed at solving drug development issues for multiple disease areas, including orphan diseases. These models and applications will vastly increase the speed and efficiency of drug development, resulting in faster, cheaper, more efficient drug trials that yield numerous new medications to ease human pain and suffering.
This work will develop Detectable Effect Cluster (DEC) analysis, a novel machine-learning based method of subgroup analysis. DEC analysis shows great promise in identifying patient subgroups with statistically significant drug effects within larger, more heterogeneous, failed therapeutic clinical trials. DEC analysis has the potential to rescue a drug that otherwise would have been discarded as a drug that does not provide therapeutic benefit, when, in fact, the opposite is true.
