A Smooth Basis for Prognostic Rules

• Activity Number: 6
• Type: Invited
• Date/Time: Sunday, August 3, 2014 : 2:00 PM to 3:50 PM
• Sponsor: WNAR
• Author(s): Michael LeBlanc*+
• Companies: Fred Hutchinson Cancer Research Center
• Keywords: regression tree ; subgroup ; prognosis ; Logic Regresion

Abstract

Survival regression trees have been widely used to describe subgroups of cancer patients with differing prognosis. We describe an extension to tree-based regression and Logic Regression that allows one to control the class of subset decision rules and the fraction of patients identified by the rules.

The proposed modeling strategy characterizes subgroups through a novel smooth basis function representation of individual ordered or continuous factors in terms of quantiles of their distribution function. To arrive at subset decision rules, we utilize extreme functions (maximum and minimum) of simple univariate transformations of the predictor variables.

Additional variance control and variable selection for subgroups is obtained through Elastic-Net type regularization. We present examples from cancer clinical trials conducted by SWOG, a national clinical trials organization.