StratifiedMedicine (version 0.2.3)

submod_otr: Subgroup Identification: Optimal Treatment Regime (through ctree)

Description

For continuous, binary, or survival outcomes, regress I(PLE>thres)~X with weights=abs(PLE) in ctree. For example, PLE could refer to individual treatment effect, E(Y|A=1,X)-E(Y|A=0, X)

Usage

submod_otr(Y, A, X, Xtest, mu_train, alpha = 0.05,
  minbucket = floor(dim(X)[1] * 0.1), maxdepth = 4, thres = ">0",
  ...)

Arguments

Y

The outcome variable. Must be numeric or survival (ex; Surv(time,cens) )

A

Treatment variable. (a=1,...A)

X

Covariate space.

Xtest

Test set

mu_train

Patient-level estimates (See PLE_models)

alpha

Significance level for variable selection (default=0.05)

minbucket

Minimum number of observations in a tree node. Default = floor( dim(train)[1]*0.05 )

maxdepth

Maximum depth of any node in the tree (default=4)

thres

Threshold for PLE, ex: I(PLE>thres). Default is ">0". Direction can be reversed and can include equality sign (ex: "<=")

...

Any additional parameters, not currently passed through.

Value

Trained ctree (optimal treatment regime) model.

• mod - tree (OTR) model object

References

Zhao et al. (2012) Estimated individualized treatment rules using outcome weighted learning. Journal of the American Statistical Association, 107(409): 1106-1118.

Examples

# NOT RUN {
library(StratifiedMedicine)

## Continuous ##
dat_ctns = generate_subgrp_data(family="gaussian")
Y = dat_ctns$Y
X = dat_ctns$X
A = dat_ctns$A

# }
# NOT RUN {
## Estimate PLEs (through Ranger) ##
res.ple = ple_train(Y, A, X, Xtest=X, family="gaussian", ple="ple_ranger")

## Fit OTR Subgroup Model ##
res_otr = submod_otr(Y, A, X, Xtest=X, mu_train = res.ple$mu_train)
plot(res_otr$mod)
# }
# NOT RUN {
# }