A0259
Title: Dynamic prediction with numerous longitudinal covariates made easy: The R package pencal
Authors: Mirko Signorelli - Leiden University, Mathematical Institute (Netherlands) [presenting]
Abstract: To make informed decisions, clinicians and patients rely on accurate predictions of the probability of experiencing adverse events (such as dementia, cancer, or death) over time. Dynamic prediction models make it possible to update the probability of experiencing an event as more longitudinal data is collected. Traditional approaches to dynamic prediction include joint modelling, which is computationally unfeasible with numerous longitudinal predictors, and landmarking, which only uses data from the last available observation. Penalized regression calibration (PRC) is introduced, a dynamic prediction method that is capable of handling numerous longitudinal covariates as predictors of survival. After illustrating the statistical methodology that PRC is based on, how the R package pencal makes it easy to estimate PRC is shown, and the predicted survival probabilities are computed and updated to validate the predicted performance of the fitted model. The results of a systematic comparison of the predictive performance of PRC and alternative modelling approaches are presented using several real-world datasets that differ in terms of survival outcome, sample size, number of longitudinal covariates, and length of the follow-up.
