hazard.ratio.plot function repeatedly estimates Cox
regression coefficients and confidence limits within time intervals.
The log hazard ratios are plotted against the mean failure/censoring
time within the interval. Unless
times is specified, the number of
time intervals will be max(round(d/e),2), where d is the
total number
of events in the sample. Efron's likelihood is used for estimating
Cox regression coefficients (using
coxph.fit). In the case of
tied failure times, some intervals may have a point in common.
hazard.ratio.plot(x, Srv, which, times=, e=30, subset,
conf.int=.95, legendloc=NULL, smooth=TRUE, pr=FALSE, pl=TRUE,
add=FALSE, ylim, cex=.5, xlab="t", ylab, antilog=FALSE, ...)
x for which to estimate hazard
ratios across time and make plots.
The default is to do so for all predictors. Whenever
one predictor is displayed, all other predictors in the
x matrix
are adjusted for (with a separate adjustment form for each time interval).
times=c(1,2,3) uses intervals
[0,1), [1,2), [2,3), [3+).
If times is omitted, uses intervals containing
e events
subset=sex=="female"
"none" for none,
"ll" for lower left of graph, or actual x and y coordinates (e.g.
c(2,3))
FALSE to suppress printing of individual Cox fits
TRUE to plot results
y-axis limits. Default is computed to include confidence bands.
x-axis, default is
"t"
y-axis, default is
"Log Hazard Ratio" or
"Hazard Ratio",
depending on
antilog.
FALSE. Set to
TRUE to plot anti-log, i.e., hazard ratio.
Frank Harrell
Department of Biostatistics, Vanderbilt University
f.harrell@vanderbilt.edu
n <- 500 set.seed(1) age <- 50 + 12*rnorm(n) cens <- 15*runif(n) h <- .02*exp(.04*(age-50)) d.time <- -log(runif(n))/h label(d.time) <- 'Follow-up Time' e <- ifelse(d.time <= cens,1,0) d.time <- pmin(d.time, cens) units(d.time) <- "Year" hazard.ratio.plot(age, Surv(d.time,e), e=20, legendloc='ll')