survival (version 3.1-8)

survobrien: O'Brien's Test for Association of a Single Variable with Survival

Description

Peter O'Brien's test for association of a single variable with survival This test is proposed in Biometrics, June 1978.

Usage

survobrien(formula, data, subset, na.action, transform)

Arguments

formula

a valid formula for a cox model.

data

a data.frame in which to interpret the variables named in the formula, or in the subset and the weights argument.

subset

expression indicating which subset of the rows of data should be used in the fit. All observations are included by default.

na.action

a missing-data filter function. This is applied to the model.frame after any subset argument has been used. Default is options()\$na.action.

transform

the transformation function to be applied at each time point. The default is O'Brien's suggestion logit(tr) where tr = (rank(x)- 1/2)/ length(x) is the rank shifted to the range 0-1 and logit(x) = log(x/(1-x)) is the logit transform.

Value

a new data frame. The response variables will be column names returned by the Surv function, i.e., "time" and "status" for simple survival data, or "start", "stop", "status" for counting process data. Each individual event time is identified by the value of the variable .strata.. Other variables retain their original names. If a predictor variable is a factor or is protected with I(), it is retained as is. Other predictor variables have been replaced with time-dependent logit scores.

The new data frame will have many more rows that the original data, approximately the original number of rows * number of deaths/2.

Method

A time-dependent cox model can now be fit to the new data. The univariate statistic, as originally proposed, is equivalent to single variable score tests from the time-dependent model. This equivalence is the rationale for using the time dependent model as a multivariate extension of the original paper.

In O'Brien's method, the x variables are re-ranked at each death time. A simpler method, proposed by Prentice, ranks the data only once at the start. The results are usually similar.

References

O'Brien, Peter, "A Nonparametric Test for Association with Censored Data", Biometrics 34: 243-250, 1978.

See Also

survdiff

Examples

# NOT RUN {
xx <- survobrien(Surv(futime, fustat) ~ age + factor(rx) + I(ecog.ps), 
			       data=ovarian) 
coxph(Surv(time, status) ~ age + strata(.strata.), data=xx) 
# }