An Interactive Dashboard for Dynamic Individualized Prediction Models for Times-to-Event in ALS Clinical Milestones

Document Type

Conference Proceeding

Publication Date

10-6-2025

Publication Title

Muscle Nerve

Abstract

Background: For many individuals with ALS, several clinically relevant events occur prior to death. The times to these clinically relevant events are valuable both for modeling disease progression and for personal planning. However, at present, there is limited ability to obtain predictions for a particular set of patient characteristics that incorporates ongoing disease progression. We developed dynamic individualized prediction models for time-to-events for several outcomes and incorporated them into a publicly available application that can aid in clinical guidance and planning. Methods: Longitudinal data from 2121 participants in the ALS Natural History Consortium dataset were used to implement landmark time-to-event analysis. Five outcomes were considered: loss of ambulation, loss of useful speech, gastrostomy, non-invasive ventilation (NIV) usage, and continuous NIV usage. In the models for each outcome, the time-varying ALSFRS-R values at the landmark time (“s”) are treated as fixed covariates in a Cox regression model from s onward. Six landmark times, between date of diagnosis and 3 years, were implemented. Covariates included age at diagnosis, sex, diagnostic delay, onset location, riluzole use, ALSFRS-R scores at the landmark time, and ALSFRS-R rates of change from baseline. Shiny™ was used to implement and present the modeling results in a freely accessible online interactive dashboard. Results: Our Shiny application allows the user to specify an outcome, landmark time, and specific covariate values (e.g., ALSFRS-R scores, onset location, etc.) to obtain predictions. The application uses these inputs to automatically produce a time-to-event prediction curve for the selected outcome using the fitted models. Additionally, the application presents risk prediction intervals for each outcome and landmark time to illustrate how covariates affect risk. Discussion: We created an interactive dashboard that leverages a dynamic landmark modeling approach to illuminate how risk for outcomes changes across time, based on a number of inputs. Our application enables users to explore risk profiles and predicted trajectories for specified sets of characteristics, providing a valuable resource for both clinical use and for those living with ALS. The application will be continuously updated and additional modeling variables, such as biomarkers like NfL, will be incorporated as they become available.

Volume

72

First Page

S97

Find It @ Sladen

Share

COinS