plotting confidence intervals for coxph using ggsurvplot

[kharknes]

Expected behavior

Using the example provided here (http://www.sthda.com/english/wiki/cox-proportional-hazards-model), I have been trying to plot two strata with confidence intervals after running a coxph. This is the example code, with data=lung added to the ggsurvplot as that was indicated as a needed fix:

data("lung")
res.cox <- coxph(Surv(time, status) ~ sex, data = lung)
res.cox
res.cox <- coxph(Surv(time, status) ~ age + sex + ph.ecog, data = lung)
summary(res.cox)
ggsurvplot(survfit(res.cox), data=lung, color = "#2E9FDF",
ggtheme = theme_minimal())
sex_df <- with(lung,
data.frame(sex = c(1, 2),
age = rep(mean(age, na.rm = TRUE), 2),
ph.ecog = c(1, 1)
)
)
sex_df
fit <- survfit(res.cox, newdata = sex_df)
ggsurvplot(fit, data=lung, conf.int = TRUE, legend.labs=c("Sex=1", "Sex=2"),
ggtheme = theme_minimal())

It should produce a nice graph like this:

Actual behavior

However, the code is producing this, with the confidence intervals all bunched up at the top:

When digging into what the survfit function is doing, it appears that the error is located there:

res.sum <- surv_summary(fit)
head(res.sum)

time n.risk n.event n.censor surv std.err upper lower strata
1 5 227 1 0 0.9949674 0.005051885 1 0.9851643 1
2 11 226 3 0 0.9799013 0.005051885 1 0.9851643 1
3 12 223 1 0 0.9748685 0.005051885 1 0.9851643 1
4 13 222 2 0 0.9647492 0.005051885 1 0.9851643 1
5 15 220 1 0 0.9596720 0.005051885 1 0.9851643 1
6 26 219 1 0 0.9546036 0.005051885 1 0.9851643 1

It then passes on a uniform confidence interval instead of one that changes by step. Is there a fix for this? I would really like to be able to graph confidence intervals for the coxph.

Steps to reproduce the problem

session_info()

# please paste here the result of
devtools::session_info()

[camposfa]

I have the same issue. It appears to be a bug introduced in version 0.4.4 (2019-05-21), as reverting to version 0.4.3 fixed the problem for me.

[stefani-odonoghue]

I reverted to version 0.4.3, but this states the R version 3.6 and 3.53 are not compatible. What R source version are you using?

[camposfa]

I'm using R 3.6.0. Try using:

require(devtools)
devtools::install_version("survminer", version = "0.4.3", repos = "http://cran.us.r-project.org")

And then restart your R session.

[kassambara]

Hi,

Thank you for pointing this issue out.

A bug has been introduced by the pull request #335

In this pull request, ifelse() has been used to test if the upper and the lower conf.int are null :

 res <- cbind(res, surv = .flat(x$surv),
                   std.err = ifelse(is.null(x$std.err), NA, .flat(x$std.err)),
                   upper = ifelse(is.null(x$upper), NA, .flat(x$upper)),
                   lower = ifelse(is.null(x$lower), NA, .flat(x$lower)))

The documentation of ifelese states that :

ifelse returns a value with the same shape as test which is filled with elements selected from either yes or no depending on whether the element of test is TRUE or FALSE.

So, since we are passing test values of length 1, we are getting results of length 1, leading to a uniform confidence interval for all rows.

I'm fixing that

[kassambara]

Fixed now, thanks.

Install the latest dev version as follow:

devtools::install_github("kassambara/survminer", build_vignettes = FALSE)

[olgasparyan]

Hi! I installed the last package version provided in here. But I still have a flat confidence interval at the value=1. Thanks!