Abstract: Joint models have been proposed to compute individual dynamic predictions from repeated measures to one or two markers. However, they hardly extend to the case where the complete patient history includes much more repeated markers. Our objective was thus to propose a solution for the dynamic prediction of a health event that may exploit repeated measures of a possibly large number of endogenous markers. We extended the random survival forest methodology to incorporate multivariate longitudinal endogenous markers. At each split of the nodes of the random forest trees, mixed models for the longitudinal markers are fitted and the predicted random effects are used among the others time-fixed predictors to split the subjects. The individual-specific event prediction is derived as the average over all trees of the leaf-specific cumulative incidence function computed using the Aalen-Johansen estimator. We demonstrate in a simulation study the performances of our methodology, both in a small and a large dimensional context. The method is applied to predict the individual risk of dementia in the elderly (accounting for the competing death) according to the trajectories of cognitive functions, brain imaging markers, and general clinical evaluation. Our method is implemented in the R package DynForest.

Biography: Robin Genuer is an associate professor in statistics at the School of Public Health (ISPED) of Bordeaux University. He defended a PhD about random forests and then worked on different aspects related to random forests (variable selection, behavior in the big data context, extension to longitudinal data and for the dynamic prediction problem) usually in an high-dimensional context with applications in health data analyses. He wrote, together with Jean-Michel Poggi, the book untitled « Random Forests with R » and its associated French version, and is the maintainer of the VSURF R package.

Please join via Zoom.

 

Meeting ID: 870 1171 6886

Code : 422654