Software Platform to Stratify Patients for Treatment Arm Randomization in Human Clinical Trials using Patient Level Predictive Models

ABSTRACT Amyotrophic lateral sclerosis ALS is a progressive neurodegenerative disease of brain and spinal cord motor neurons Since Riluzole was approved in over late phase clinical trials have failed and no additional medications have been approved for ALS While death from ALS averages to years from onset of symptoms disease progression displays wide heterogeneity A typical year long clinical trial can have to of the patients die from the disease while another similar percentage progresses very slowly if at all Several meta analyses of ALS trial data indicate that ALS clinical trials are prone to statistical uncertainty and would benefit from tools that increase statistical sensitivity It is clear that current statistical tools are inadequate to address the drug development challenges posed by this disease and many other diseases that characteristically exhibit high heterogeneity in disease progression Using the recently available ALS PRO ACT data set our team was recently declared a winner of the DREAM Phil Bowen ALS Prediction Prize Life Challenge Since the contest we have significantly improved the algorithm built several additional models and begun to create drug development tools The goal of this grant is to validate our clinical trial randomization tool and develop a prototype interface for use and testing at clinical trial sites This prototype will serve as a platform for building a suite of tools based on disease progression predictions of individual patients that will eventually be used for drug development in multiple indications The key innovation of this work as it applies to randomizing patients for inclusion in different arms of a clinical trial is that it stratifies patients not by a set of features at the beginning of a trial but rather by predicted outcome at the end of the trial as if patients in the treatment arm had not received the intervention being tested An improved trial arm balance will provide a better test of the efficacy of the intervention This work will focus on the following Specific Aims Aim Demonstrate that compared to traditional randomization strata randomization strata defined by predictive algorithms significantly improve the balance of outcome features at the end of a trial period Aim Work with our clinical partner to develop a prototype platform that will enable an ALS predictive algorithm to be used by on site investigators for randomization in future clinical trials The randomization tool is the first in a series of planned tools based on patient level disease progression predictions These tools will radically change the way early ALS clinical trials are enrolled simulated and analyzed and will enable the development of similar tools not only for other neurodegenerative diseases such as Parkinson s and Alzheimer s but also for multiple other diseases including diabetes hospital acquired infections heart disease and cancer NARRATIVE This work will develop a prototype to test the use of patient disease progression predictions made by machine learning models as a new way of randomizing clinical trials The prototype will serve as a platform for the inclusion of a range of drug development tools based on individual patient predictions