Abstract
In survival analysis, dynamic prediction refers to the process of updating the predicted probability that an individual will experience an event in the future, each time new repeated measurement data become available. Traditionally, this was primarily accomplished through simple landmarking models or computationally intensive joint modelling methods. Since 2019, there has been a surge in methodological work aimed at making more efficient use of available longitudinal covariate data. In this talk, I will take attendees on a journey through some of these recent developments. Along the way, we will make three stops on the itinerary: first, a critical look at which types of data should - and shouldn't - be used to train dynamic prediction models [1]; second, a new statistical method and open-source software designed to deal with settings with numerous longitudinal covariates [2,3]; and finally, real-world applications that can help us understand the strengths and limitations of these emerging approaches.
Link Microsoft Teams
Organization:
Saverio Ranciati