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

Background and objectives: Times to clinically relevant events are a valuable outcome in observational and interventional studies, complementing linear outcomes such as functional rating scales and biomarkers. In ALS, there are several clinically relevant events. We developed dynamic prediction models for several of these times to events that can be used for clinical trial modeling and personal planning.

Methods: Landmark time-to-event analysis was implemented to determine the effect of patient characteristics on disease progression. Longitudinal data from 1557 participants in the ALS Natural History Consortium dataset were used. Five outcomes in the ALS disease progression were considered: loss of ambulation, loss of speech, gastrostomy, noninvasive ventilation (NIV) use, and continuous NIV use. Covariates in our models include age at diagnosis, sex, onset location, riluzole use, diagnostic delay, ALSFRS-R scores at the landmark time, and ALSFRS-R rates of change from baseline. Internal and external validation techniques were used.

Results: For each of our models and landmark times, we present risk prediction intervals for random sets of patient characteristics. We demonstrate our models' application for an individual's personal predicted time-to-event. Our internal and external validation metrics indicate good concordance and overall performance. The time to loss of speech models perform the best for each metric in terms of both internal and external validation.

Discussion: Landmarking is an efficient, individualized risk prediction model that is intuitive for both clinicians and patients. Importantly, landmarking can be used for clinical trial modeling, personal planning, and development of real-world evidence of the impacts of treatment interventions.