In this paper we present prediction and variable importance (VIM) methods for longitudinal data sets containing both continuous and binary exposures subject to missingness. We demonstrate the use of these methods for prognosis of medical outcomes of severe trauma patients, a field in which current medical practice involves rules of thumb and scoring methods that only use a few variables and ignore the dynamic and high-dimensional nature of trauma recovery. Well-principled prediction and VIM methods can thus provide a tool to make care decisions informed by the high-dimensional patient’s physiological and clinical history. Our VIM parameters can be causally interpreted (under causal and statistical assumptions) as the expected outcome under time-specific clinical interventions. The targeted MLE used is doubly robust and locally efficient. The prediction method, super learner, is an ensemble learner that finds a linear combination of a list of user-given algorithms and is asymptotically equivalent to the oracle selector. The results of the analysis show effects whose size and significance would have been not been found using a naive parametric approach, as well as improvements of up to 0.07 in the AUROC.
Diaz, Ivan; Hubbard, Alan E.; Decker, Anna; and Cohen, Mitchell, "Variable Importance and Prediction Methods for Longitudinal Problems with Missing Variables" (October 2013). U.C. Berkeley Division of Biostatistics Working Paper Series. Working Paper 318.