paf (version 1.0)

paf: Calculate attributable fraction function for censored survival data

Description

Fit a Cox model and calculate the unadjusted/adjusted attributable fraction function of a set of covariates based on the Cox model using the method proposed by Chen, Lin and Zeng (2010).

Usage

paf(formula, data, cov)

Arguments

formula

a formula object for the Cox model considered , which has the same format as that in the coxph function of the survival package.

data

a data.frame in which to interpret the variables named in the formula.

cov

the set of covariates whose attributable fraction function is of interest.

Value

time

unique uncensored event times at which the attributable fraction function jumps.

est

the estimates of unadjusted/adjusted attributable fractions at unique uncensored event times.

se

the standard errors of the estimated attributable fractions.

low

the lower confidence limits of the atrtributable fractions.

upp

the upper confidence limits of the atrtributable fractions.

fit.cox

coxph object from the fitted Cox model.

Details

This function calculates the unadjusted/adjusted attributable fraction function for the set of covariates specified in cov which must also be included as covariates of the Cox model. The function calculates the unadjusted attributable fraction function if the Cox model does not include other covariates; otherwise the function calculates the adjusted attributable fraction function adjusting for other covariates in the Cox model.

References

Chen L, Lin DY, Zeng D. (2010). Attributable fraction functions for censored event times. Biometrika 97, 713-726.

See Also

plot.paf.

Examples

# simulated data set from a Cox model
n = 1000
x1 = as.numeric(runif(n)>0.5)
x2 = x1 + rnorm(n)
t = exp(-x1 - 0.5 * x2) * rexp(n, rate = 0.1)
c = runif(n, 0, 3.4)
y = pmin(t, c)
delta = as.numeric(t<=c)
test = data.frame(time=y, status=delta, x1=x1, x2=x2)

# calculate the atrributable fraction function of x1 adjusting for x2
result=paf(Surv(time, status) ~ x1 + x2, data=test, cov=c('x1'))
result$fit.cox
cbind(result$time, result$est, result$se, result$low, result$upp)[1:10, ]
# Calculate the unadjusted attributable fraciton function of x1
result=paf(Surv(time, status) ~ x1, data=test, cov=c('x1'))