Title: Dynamic survival prediction for multivariate joint models using the R package joineRML
Authors: Graeme Hickey - University of Liverpool (United Kingdom) [presenting]
Pete Philipson - University of Northumbria (United Kingdom)
Andrea Jorgensen - University of Liverpool (United Kingdom)
Ruwanthi Kolamunnage-Dona - University of Liverpool (United Kingdom)
Abstract: Methods for the joint analysis of time-to-event data and longitudinal data have been developed in recent years, with most emphasis on modelling and estimation. Moreover, research has predominantly concentrated on the joint modelling of a single longitudinal outcome. In clinical practice, the data collected might be more complex, featuring multiple longitudinal outcomes. Harnessing all available measurements in a single model is advantageous and should lead to improved inference and more specific model predictions. In recent years there has been a growing interest in the application of prognostic models to the field of personalised medicine, which can be leveraged by clinicians to adapt care optimally to patients anticipated to deteriorate. Here we focus on the dynamic prediction of a subjects failure time conditional on their observed history of multivariate longitudinal outcome measurements. We will explore the influence of moving from a univariate dynamic prediction to a multivariate framework. We describe an R package -- joineRML -- recently available on CRAN (https://cran.r-project.org/web/packages/joineRML/index.html), which fits joint models to time-to-event data and multivariate longitudinal data. We demonstrate the latest extension of this package for calculating out-of-data dynamic predictions. The package and methodology are illustrated using a real-world clinical dataset that records several repeatedly measured biomarkers.