Model-Based and Model-Free Techniques for Amyotrophic Lateral Sclerosis Diagnostic Prediction and Patient Clustering

Area of Science:

• Neuroscience
• Data Science
• Biostatistics

Background:

• Amyotrophic lateral sclerosis (ALS) is a progressive neurodegenerative disease affecting approximately 5 in 100,000 individuals in the US.
• Accurate forecasting of ALS progression and patient phenotyping are crucial for clinical decision support.
• Existing data analysis methods have limitations in predicting individual patient trajectories.

Purpose of the Study:

• To develop and compare machine learning models for predicting ALS disease progression, measured by changes in the Amyotrophic Lateral Sclerosis Functional Rating Scale (ALSFRS) score.
• To employ unsupervised learning for reliable and reproducible clustering of ALS patients into distinct computable phenotypes.
• To leverage the PRO-ACT database for robust data-driven insights into ALS patient heterogeneity.

Main Methods:

• Utilized a large ALS dataset (8,000 patients, 3 million records) from the PRO-ACT archive.
• Applied both model-based (linear models) and model-free (random forest, BART) machine learning techniques for outcome prediction.
• Employed unsupervised machine learning for patient cohort clustering and computable phenotype derivation.

Main Results:

• Predicting individual ALSFRS score changes yielded moderate success (correlation coefficients up to 0.545).
• Unsupervised clustering reliably stratified patients into four distinct computable phenotypic subgroups with over 95% consistency.
• Salient clinical features from the PRO-ACT archive were identified to explicate the derived patient subcohorts.

Conclusions:

• While direct prediction of univariate clinical outcomes in ALS remains challenging, data science strategies effectively cluster patients.
• Unsupervised clustering generates stable and reliable computable phenotypes, offering valuable insights into ALS patient heterogeneity.
• These findings support the generation of evidence-based hypotheses regarding complex, multivariate factors influencing ALS progression.