^Classification And Regression Tree analysis^ ^-------------------------------------------^
^cart^ varlist [^if^] [^in^] ^, t^ime^(^var^) f^ail^(^var^)^ [ ^s^trata^(^varlist^) a^djust^(^varlist^) na^me^(^string^)^ ^p^val^(^real 0.05^) mins^ize^(^int 10^) minf^ail^(^int 10^)^ ^su^mby^(^varlist^) tab^by^(^varlist^)^ ^at(^string^)^]
^Description^ ^-----------^ This program performs a CART analysis for failure time data. It uses the martingale residuals of a Cox model to calculate (approximate) chisquare values for all possible cutpoints on all the CART covariates.
^Varlist^ ^-------^ The varlist (required with minimum one variable) specifies the independent variables for CART. Observations with missing values on any of the variables in this list are deleted.
^time()^ specifies the time variable. Observations with values <0 or missing are deleted. Value 0 is replaced by a small positive value. ^fail()^ specifies the failure indicator. Observations with values <0 or missi > ng are deleted and values >0 are considered failures. ^strata()^ Optional. Specifies a varlist for a priori stratification of Cox mo > del ^adjust()^ Optional. Specifies a varlist for apriori adjustment of Cox model. Default the null Cox model will be used for calculation of the individual relative hazard rate = exp(b'x), i.e. =1 in case of the null model, which will form the basis for the calculation of the expected amount of events for each subject within a CART subgroup. However it is also possible to calculate the expected number after adjustment for some variables and stratification by others. This gives the opportunity to perform a CART for some variables after adjustment and/or stratification by other variables in a Cox model. The Cox model for adjustment and the associated exp(b'x) is fitted only once, but the expected number of failures per subject is recalculated within each CART subgroup. ^pval()^ specifies the maximum P-value for a split; default ^pval^=0.05. P-values will be adjusted for the search of the optimal cutpoint within a range within a group by the formulas of Miller&Siegmund and Worsley. A split is not allowed if the adjusted minimum P-value is larger than ^pva > l^. See Lausen for more details. ^minsize()^ specifies the minimum size of a subgroup for a split; default 10. ^minfail()^ specifies the minimum number of failures in a subgroup that is required before a split will be attempted. Default 10. ^Output control:^
^sumby()^ specifies varlist for a summary of these variables by the final CART grouping variable.
^tabby()^ specifies varlist for a tabulation of these variables by the final C > ART grouping variable.
^at()^ specifies timepoints for display of survival probabilities by the final CART grouping variable. See ^help sby^ for the use of the ^at^ option. ^name()^ specifies a name (default _cart) that will be used for - the log-file - the cart-tree graph - the dta file with results - the CART grouping variable. The _cart splitting history is posted to a file with name <name>x, default > _cartx.
^References^ ^----------^ Lausen et al, in Computational Statistics (Eds. P Dirschedl, R Ostermann),p 483 > -496, 1994. Lausen et al, Informatik, Biometrie und Epidemiologie in Medizin und Biologie 2 > 8, 1-13, 1997 Miller and Siegmund, Biometrics 38, 1011-1016, 1982. Worsley, Technometrics 25, 35-42, 1983.
^Author:^ ^-------^ Wim van Putten Erasmus MC - Daniel den Hoed Cancer Center Rotterdam The Netherlands FAX: +31.10.4391028 e-mail: w.vanputten@@erasmusmc.nl