ABSTRACT
In the past two decades, much research has focused on the terminal decline hypothesis of cognitive performance in very old age. Only in recent years, however, has this hypothesis been assessed with advanced methodologies. One such proposal used the joint longitudinal+survival model to (a) assess cognitive performance and change therein, across a wide set of cognitive tasks, and (b) assess how cognitive performance and particularly change therein could predict survival in a sample of older individuals (Ghisletta, McArdle, & Lindenberger, 2006). In this chapter we expand on that inquiry by making use of recent data mining techniques that more aptly consider the concomitant effects of a large number of predictors and rely on resampling techniques to achieve robust results. Prediction survival trees and random survival forests are used to study the effect of cognition on survival in the Berlin Aging Study.
