ALS Survival Prediction Model with Advanced-Stage Patient Data

A machine learning model trained on 7,974 ALS patient records from the PRO-ACT database and a supplemental cohort of 678 advanced-stage patients. The gradient-boosting model achieved a C-index of 0.709 for survival prediction and successfully stratified patients into low-, average-, and high-risk tertiles. The model and supporting data were published by Danielle Beaulieu on figshare in April 2026.

Use Cases

• Training survival prediction models based on baseline patient characteristics like vital capacity and ALSFRS-R scores.
• Stratifying ALS patients for clinical trial enrollment based on predicted survival risk tertiles.
• Analyzing the relative importance of clinical parameters such as VC% slope and age for driving survival outcomes.

Strengths

• Model was validated on a distinct external cohort of 403 advanced-stage ALS patients.
• Reports specific performance metrics, including a C-index of 0.709 and calibration statistics.
• Focus includes a population often excluded from studies: patients with ≤50% predicted normal vital capacity.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the underlying training data is unknown, which may limit suitability assessment.
• The dataset is small at 855.3 KB, indicating it likely contains model documentation rather than the full raw patient data.

Provenance

Source

PRO-ACT Database supplemented with advanced-stage patient cohorts.

Collection Method

Training data was supplemented with advanced-stage ALS patients for model development.

Freshness

Last updated 2026-04-13 20:59:54; freshness should be verified.