A machine-learning based objective measure for ALS disease severity.

Area of Science:

• Neurology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Amyotrophic Lateral Sclerosis (ALS) disease severity assessment traditionally relies on the subjective ALS Functional Rating Scale (ALSFRS-R).
• Objective measures are crucial for robust evaluation of drug effectiveness in clinical trials and real-world settings.
• Developing objective tools can enhance participant identification for cohort studies and provide more sensitive disease tracking.

Purpose of the Study:

• To develop and validate machine learning (ML) models for objective measurement of ALS disease severity.
• To utilize voice and accelerometer data for predicting bulbar and limb-related ALSFRS-R scores.
• To assess the utility of ML-derived scores in evaluating drug effects and capturing subtle disease progression.

Main Methods:

• Longitudinal data from 584 individuals with ALS were collected over four years, including voice samples and accelerometer measurements.
• ML models were trained to predict ALSFRS-R scores using voice (predicting bulbar function) and accelerometer (predicting limb function) data.
• Model performance was evaluated on a test set (n=109), and ML-derived scores were compared with self-reported ALSFRS-R and drug effects (edaravone).

Main Results:

• Voice-based ML models achieved an AUC of 0.86 for speech-related ALSFRS-R prediction.
• Accelerometer-based ML models achieved a median AUC of 0.73 for limb-related function prediction.
• ML-derived scores preserved correlations observed in self-reported scores and showed consistency with subjective measures when evaluating edaravone's effects. Continuous ML scores captured gradual changes missed by integer ALSFRS-R scores.

Conclusions:

• ML models trained on voice and accelerometer data provide objective and sensitive measures of ALS disease severity.
• These objective measures are valuable for assessing drug efficacy, tracking disease progression, and potentially improving clinical trial design.
• The developed ML tools offer a promising advancement over subjective rating scales for ALS management and research.