Predicting functional decline and survival in amyotrophic lateral sclerosis

Area of Science:

• Biomedical Informatics
• Neuroscience
• Clinical Trial Design

Background:

• Amyotrophic lateral sclerosis (ALS) disease course prediction is crucial for optimizing clinical trials.
• The Prize for Life foundation established the PRO-ACT database, containing de-identified ALS patient data from clinical trials.
• PRO-ACT data enables researchers to develop predictive models for ALS progression.

Purpose of the Study:

• To develop predictive models for amyotrophic lateral sclerosis (ALS) functional decline and survival.
• To identify key variables that can accurately predict disease course.
• To inform the design of more targeted and efficient ALS clinical trials.

Main Methods:

• Time series data from the PRO-ACT database were fitted to exponential models.
• Machine learning algorithms were employed to predict functional decline (fast/slow) and survival risk (high/low).
• Model performance was evaluated using cross-validation, Receiver Operator Curves (AUC), and root mean squared errors.

Main Results:

• A boosting algorithm model using post-baseline changes in weight, alkaline phosphatase, albumin, and creatine kinase predicted functional decline class with AUC = 0.82.
• Predicting functional decline based on baseline characteristics alone was unsuccessful.
• Baseline levels of total bilirubin, gamma glutamyltransferase, urine specific gravity, and ALSFRS-R item score (climbing stairs) accurately predicted survival class.

Conclusions:

• Predictive models utilizing a small number of variables can accurately classify functional decline and survival in ALS patients within a 1-2 year timeframe.
• These models hold potential for improving the efficiency and targeting of future amyotrophic lateral sclerosis clinical trials.
• Biomarker combinations show promise for predicting ALS disease trajectory.