生存函数图中的哪条曲线是哪条曲线？

Question

I am plotting survival functions with the survival package. Everything works fine, but how do I know which curve is which? And how can I add it to a legend?

  url <- "http://socserv.mcmaster.ca/jfox/Books/Companion/data/Rossi.txt"
  Rossi <- read.table(url, header=TRUE)[,c(1:10)]
  km <- survfit(Surv(week, arrest)~race, data=Rossi)
  plot(km, lty=c(1 ,2))

Answer 1

how do I know which curve is which?

Using str() you can see which elements are in km . km$strata shows there are 48 and 10 elements. This coincides with the declining pattern of the first 48 items and last 10 items in km$surv

km$surv[1:48]
km$surv[49:58]

So in addition to the hint on order in print() , with this particular dataset we can also be sure that the first 48 elements belong to race=black

And how can I add it to a legend?

Unlike other model output km is not easily transformed to a data.frame. However, we can extract the elements ourselves and create a data.frame and then plot it ourselves.

First we create a factor referring to the strata: 48 blacks and 10 others

race <- as.factor(c(rep("black", 48), rep("other", 10)))
df <- data.frame(surv = km$surv, race = race, time = km$time)

Next we can plot it as usual (in my case, using ggplot2).

library(ggplot2)
ggplot(data = df, aes(x = time, y = surv)) + 
    geom_point(aes(colour = race)) + 
    geom_line(aes(colour = race)) +
    theme_bw()

Answer 2

Unfortunately, the plot.survival function doesn't seem to offer a nice option for labeling the curves. The documentation says that curves are plotted in the order they appear in print , so you can figure out which is which when you vary line type or color. But that's not great for sharing.

One alternative is to use the survplot function from rms , which does label the curves. Here's how that looks with your example and CI plotting turned off. (Note that survplot won't take a survfit object, so you've got to redo the estimation with a function whose results it can read -- here, npsurv .)

library(rms)
survplot(npsurv(Surv(week, arrest)~race, data=Rossi), conf = "none")

Check the documentation for ways to tweak other aspects of the chart, including replacing the labels in the plot with a legend.

Answer 3

Thanks to the answer of Richard, I found a way to plot the right names to the right curves, with the base plot way that is used in plot.survfit :

legend_values <- names(km$strata)
plot(km)
legend(
  "topright",
  legend=legend_values,
  col=1:2,
  lty = c(1,1),
  horiz=FALSE,
  bty='n')

I prefer the ggplot way of plotting, but I like to retain the stepwise presentation in plot.survfit .