Abstract:
|
Subgroup identification for personalized medicine become very popular in the last decade. Efficient tree-based methods adapted from machine learning are natural candidates since they provide subgroups as terminal nodes. However additional challenges arises when evaluating the type I error rates and selection bias for estimated treatment effect in subgroups, which becomes a standard practice for any data-driven subgroup analysis with clinical data. This can be achieved by computationally intensive resampling methods hence the importance of using efficient recursive partitioning algorithms for overcoming these challenges. In this talk we consider computational aspects of a recursive partitioning-based SIDES method (Lipkovich et al. 2011), specifically the fast evaluation of cut-offs for candidate covariates and computing adjustments for selection bias for a variety of splitting criteria and clinical outcomes as well as parallel implementation when conducting data resampling.
|