GUIDE is a multi-purpose algorithm for classification and regression tree construction with special capabilities for identifying subgroups with differential treatment effects. It is unique among subgroup methods in having all these features: unbiased split variable selection, approximately unbiased estimation of subgroup treatment effects, treatments with two or more levels, allowance for linear effects of prognostic variables within subgroups, and automatic handling of missing pre- dictor variable values without imputation in piecewise-constant models. Predictor variables may be continuous, ordinal, nominal, or cyclical (such as angular mea- surements, hour of day, day of week, or month of year). Response variables may be univariate, multivariate, longitudinal, or right-censored. This article gives a current account of the main features of the method for subgroup identification and reviews a bootstrap method for conducting post-selection inference on the subgroup treatment effects. A data set pooled from studies of amyotrophic lateral sclerosis is used for illustration. Key
CITATION STYLE
Loh, W.-Y., & Zhou, P. (2020). The GUIDE Approach to Subgroup Identification (pp. 147–165). https://doi.org/10.1007/978-3-030-40105-4_6
Mendeley helps you to discover research relevant for your work.