Online Program Home
My Program

Abstract Details

Activity Number: 254 - Contributed Poster Presentations: Section on Statistical Learning and Data Science
Type: Contributed
Date/Time: Monday, July 30, 2018 : 2:00 PM to 3:50 PM
Sponsor: Section on Statistical Learning and Data Science
Abstract #329307
Title: Machine-Learning Approach to Defining Covariates to Increase Study Power in ALS Clinical Trials and Other Multifactorial Heterogeneous Disease Areas
Author(s): Danielle Beaulieu* and Albert Taylor and Samad Jahandideh and David Ennist and Andrew Conklin and Mike Keymer
Companies: Origent Data Sciences and Origent Data Sciences and Origent Data Sciences and Origent Data Sciences and Origent Data Sciences and Origent Data Sciences
Keywords: machine-learning; covariates; clinical trials; power; ALS
Abstract:

Defining baseline characteristics for covariate-adjusted analyses to increase study power is not new. However, highly multifactorial heterogeneous diseases, such as Amyotrophic Lateral Sclerosis (ALS), present a challenge in defining few baseline covariates predictive of outcomes to add substantial benefit to study power. Thus, we developed non-linear, non-parametric, machine-learning (ML) models that utilize the full breadth of available patient data, between 15 and 30 baseline features, and provide a single prediction value for each disease outcome of interest that can be used as covariates in the analysis. For ALS, we have used the PRO-ACT database to train and internally validate ML models that predict trial endpoints of survival, change in ALSFRS-R, and change in % expected vital capacity. Simulations were performed using the PRO-ACT population and predictions from 10-fold cross validation that showed potential increases in study power > 10% using our predictions as covariates compared to unadjusted or traditional covariate-adjusted analyses using one or two baseline characteristics. Preliminary modeling in other neurodegenerative diseases show similarly promising results.


Authors who are presenting talks have a * after their name.

Back to the full JSM 2018 program