Abstract: The individual data collected throughout patient follow-up constitute crucial information for assessing the risk of a clinical event. 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. We extended the random survival forest methodology to incorporate multivariate longitudinal endogenous markers. The random survival forest is composed by an ensemble of decision trees, where the subjects are recursively split into two subgroups. At each split, 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. We demonstrate in a simulation study the performances of our methodology. We also applied it to predict the individual risk of dementia in the elderly according to the trajectories of cognitive functions, brain imaging markers, and general clinical evaluation. Our random survival forest extends the joint modelling methodology to predict clinical events from individual longitudinal history when the number of repeated markers is large.

Keywords: Individual prediction; Joint modeling; Random survival forests; Longitudinal modeling

Ishwaran H, Kogalur UB, Blackstone EH and Lauer MS (2008). Random survival forests. The Annals of Applied Statistics, 2(3), 841-860.
Laird NM and Ware JH (1982). Random-Effects Models for Longitudinal Data. Biometrics, 38(4), 963-974.