B1365
Title: Pencal: An R package for the dynamic prediction of survival with many longitudinal predictors
Authors: Mirko Signorelli - Leiden University, Mathematical Institute (Netherlands) [presenting]
Abstract: Longitudinal and high-dimensional measurements are increasingly common in biomedical research. Repeated measurement data carry important information about ageing and disease progression that can be used to update predictions of survival outcomes dynamically. Despite the availability of several methods to predict survival from either a handful of longitudinal covariates or a high-dimensional set of cross-sectional covariates, until recently, methods that could deal with a large number of longitudinal covariates were missing. The aim is to introduce penalized regression calibration, a new method to predict survival using a large (potentially high-dimensional) number of longitudinally-measured covariates as predictors, and the R package pencal that has been designed to make it easy for users to estimate PRC and use it for dynamic prediction. The problem of obtaining unbiased estimates of predictive performance is discussed, and how pencal exploits parallelization to compute them efficiently.