Abstract Details

Title Machine Learning Enables Smaller ALS Clinical Trials

Topic Neuromuscular and Clinical Neurophysiology (EMG)

Presentation(s) P1 - Poster Session 1 (8:00 AM-9:00 AM)

Poster/Presentation
Number 11-003

Objective

Use machine learning to predict progression of Amyotrophic Lateral Sclerosis (ALS) and show how integrating predictions into clinical trial designs can reduce the sample size required to achieve desired power.

Background

ALS is a progressive neurodegenerative disease with few treatment options available. Difficult diagnosis, clinical heterogeneity, and low prevalence make recruitment for new clinical trials challenging.

Design/Methods We used the Pooled Resource Open-Access ALS Clinical Trials (PRO-ACT) dataset to train a Conditional Restricted Boltzmann Machine (CRBM) to predict ALS progression. CRBMs are a powerful class of machine learning models that learn distributions of correlated variables over time from historical data. Once trained, a CRBM generates a comprehensive longitudinal clinical prediction from a subject’s baseline characteristics, which we refer to as their Digital Twin. Digital Twins describe a subject’s likely progression under standard-of-care and include predictions for Revised Amyotrophic Lateral Sclerosis Functional Rating Scale (ALSFRS-R), Forced Vital Capacity (FVC), labs, vitals. We applied a statistical technique we developed, PRognostic COVariate Adjustment (PROCOVA), to quantify the sample size reduction achieved by adjusting for Digital Twin predictions (i.e., prognostic scores) in a clinical trial design.

Results

We evaluated Digital Twins on a held-out portion of the dataset (not used for training) and found that longitudinal predictions were consistent with observed data. Assuming ALSFRS-R change from baseline as the primary endpoint, we estimated that adjusting for Digital Twin predictions in addition to age and rate of progression (i.e., rate of ALSFRS-R decline calculated from symptom onset) can lead to a 27% reduction of the control arm compared to no adjustment and a 12% reduction compared to adjusting for age and rate of progression alone.

Conclusions Digital Twins accurately model ALS progression under standard-of-care and can be integrated into clinical trial designs to reduce the sample size required to achieve desired power.

Authors/Disclosures
Daniele Bertolini, PhD (Unlearn.AI) PRESENTER	Dr. Bertolini has received personal compensation for serving as an employee of Unlearn.AI, Inc. Dr. Bertolini has received personal compensation for serving as an employee of Quid, Inc.
	No disclosure on file
	No disclosure on file
Jonathan Walsh, PhD (Unlearn.AI)	Dr. Walsh has received personal compensation for serving as an employee of Unlearn.AI. Dr. Walsh has received intellectual property interests from a discovery or technology relating to health care.

��ɫ��

��ɫ��