Post date: Apr 4, 2016 3:55:23 PM
Title: Censoring Unbiased Survival Trees and Forests
Speaker: Jon Steingrimsson, Department of Biostatistics, JHU
[Abstract]
Survival trees use recursive partitioning to separate patients into distinct risk groups. This work extends previous inverse probability censoring weighted-loss-based recursive partitioning procedures to the case of doubly robust loss functions. The doubly robust loss functions utilize available information better and are more robust to the modeling choices made. The performance of the resulting survival trees is evaluated through simulation studies and through analyzing data on death from myocardial infraction. We further present extensions to random survival forests and discuss practical issues related to implementation of the algorithms. This is joint work with Rob Strawderman, Liqun Diao, and Annette Molinaro.