Time-to-event prediction in ALS using a landmark modeling approach, using the ALS Natural History Consortium dataset

Area of Science:

• Neurology
• Biostatistics
• Clinical Trials

Background:

• Clinically relevant time-to-event outcomes are crucial in observational and interventional studies for diseases like Amyotrophic Lateral Sclerosis (ALS).
• Several key clinical events mark ALS progression, necessitating accurate prediction for patient management and research.
• Dynamic prediction models can enhance the utility of time-to-event data in ALS research and clinical practice.

Purpose of the Study:

• To develop and validate dynamic prediction models for clinically relevant time-to-event outcomes in ALS.
• To enable personalized risk prediction for individual ALS patients.
• To provide tools for improved clinical trial modeling and patient planning in ALS.

Main Methods:

• Landmark time-to-event analysis was applied to longitudinal data from 1557 ALS participants.
• Five ALS progression events were modeled: loss of ambulation, speech, gastrostomy, noninvasive ventilation (NIV) use, and continuous NIV use.
• Models incorporated covariates such as age at diagnosis, sex, onset, riluzole use, diagnostic delay, and ALSFRS-R scores/rates of change, with internal and external validation.

Main Results:

• Dynamic prediction models were created for multiple ALS time-to-event outcomes.
• Risk prediction intervals were generated for patient subgroups and individual predictions.
• Models demonstrated good concordance and performance, with time to loss of speech models showing the best validation metrics.

Conclusions:

• Landmarking provides an efficient, individualized risk prediction method for ALS.
• These models are valuable for clinical trial design, personal patient planning, and generating real-world evidence on treatment impacts.