Individualized predictions for clinical milestone in amyotrophic lateral sclerosis: A multialgorithmic approach

Area of Science:

• Neurology
• Biostatistics
• Medical Informatics

Background:

• Amyotrophic lateral sclerosis (ALS) presents significant phenotypic heterogeneity.
• Predicting clinical milestones like loss of swallowing function is difficult for individual ALS patients.

Purpose of the Study:

• To develop individualized prediction models for estimating the time to loss of autonomy in swallowing function for ALS patients.
• To assess the performance of different time-to-event prediction algorithms for this crucial clinical milestone.

Main Methods:

• Utilized the Pooled Resource Open-Access ALS Clinical Trials (PRO-ACT) database.
• Developed and compared three models: Accelerated Failure Time (AFT), Cox Proportional Hazard (COX), and Random Survival Forest (RSF).
• Defined the target variable as the time to a decline in the ALSFRS-R swallowing score to 1 or below.

Main Results:

• All models demonstrated strong internal predictive performance with high concordance indices (C-indices) during cross-validation.
• External validation confirmed comparable discriminative power across datasets, though calibration showed a tendency for overestimation in real-world data.
• While effective at patient stratification, the Random Survival Forest model's results showed less alignment with Kaplan-Meier curves compared to AFT and COX models.

Conclusions:

• Individualized prediction models for swallowing function decline in ALS have been developed and implemented in a web application.
• These models can support individualized patient counseling, management strategies, and clinical trial design for interventions like gastrostomy.
• Further model optimization is recommended to advance personalized care for ALS patients.