Table 2.

Summary of published survival tree algorithms.

Author(s)	Splitting rule	Pruning rule	Implementation
Gordon and Olshen (1985)	Impurity (specifically defined based on KM curves) reduction	Cost-complexity pruning and cross-validation	STREE
Ciampi, Thiffault, Nakache, and Asselain (1986)	Log-rank test statistic	Akaike information criterion (AIC)	Splitting criterion implemented in STREE
Segal (1988)	Log-rank test statistic	Not available	Splitting criterion implemented in STREE
Butler, Gilpin, Gordon, and Olshen (1989) Davis and Anderson (1989)	Log-rank test statistic Exponential log-likelihood	A within-node measure Cost-complexity pruning
Therneau, Grambsch, and Fleming (1990)	Martingale residuals	Cost-complexity pruning and cross-validation
LeBlanc and Crowley (1992)	First step of full likelihood	Cost-complexity pruning and cross-validation	Splitting criterion implemented in R package “rpart;” STREE also has a slightly modified version.
LeBlanc and Crowley (1993)	Log-rank test statistic	Resampling and permutation
Intrator and Kooperberg (1995)	Log-rank test statistic	Cost-complexity pruning
Zhang and Singer (1999)	A weighted combination of impurity of the death indicator and impurity of the time	Cost-complexity pruning
Breiman (2002)	Probability .75 to split on time, and Probability .25 to split on a covariate	N/A (embedded within the survival forest algorithm)	Breiman (2003a, 2003b)
Molinaro, Dudoit, and van der Laan (2004)	An inverse probability of censoring weighted (IPCW) loss function	Cost-complexity pruning and cross-validation	Use R package “rpart” by providing IPCW weights
Hothorn et al. (2006b)	Minimum p value	Stop when no p value is below a pre-specified a-level	R package “party”