Dimension reduction and variable selection in case control studies via regularized likelihood optimization