Feature request: additional arguments for `theme_risktable_default()`

[jpnolte]

First of all, thank you so much for the amazing work. Your package is so much better than the survminer package I used before. I've done a few projects now with your package and some details in adapting the risk table would be a great addition.

Lets take this a small reprex:

p <- survfit2(Surv(time, status) ~ sex, data = df_lung) |>
  ggsurvfit() +
  add_risktable(risktable_stats = "n.risk") + 
  scale_ggsurvfit()

(1) When dealing with larger patient numbers (e.g., everything above 1000) it will usually lead to the numbers at timepoint 0 overlap with the axis text on the y axis of the risk table. In the add_risktable() documentation you can see it in example 3. So my solution up until now is to either increase the expand argument in scale_x_continouus() (e.g., scale_x_continuous(expand = c(0.015, 0))) or play with the width argument when saving with ggsave() until the text does not overlap anymore.

But what if you would include an argument called margin within theme_risktable_default() which would work like the following but for the risktable and not the plot.

p + theme(axis.text.y = element_text(margin = margin(r = 10)))

(2) Left align the axis text of the y-axis of the risk table. This is more of a subjective option, but if you are already thinking about creating the first (important) one, it might be fast to also integrate this as an additional argument (hjust or align) to theme_risktable_default(). Here is how to do it for the plot. Would be nice to have it for the risk table as well.

p + theme(axis.text.y = element_text(hjust = 0))

(3) Positioning the stats_label above the axis text of the y-axis in the risk table instead of on top of the table itself. I didn't have time yet to look up your way of defining the stats_label (e.g., At risk), but it looks like you use ggtitle(). If so, this might also be quite easy to integrate and useful since some other programs seem to support it.

p +
  ggtitle("Testing this") + 
  theme(
    plot.caption = element_text(hjust = 0),
    plot.title.position = "plot"
  )

(4) Right align text (the numbers) from geom_text() in the risk table. I know this is only a matter of appearence, but it gets way easier to read and compare larger numbers with smaller ones. This one I did not figure out yet, since providing geom_text() through add_risktable() with the hjust argument for some reason puts a small offset there additionally.

survfit2(Surv(time, status) ~ sex, data = df_lung) |>
  ggsurvfit() +
  add_risktable(risktable_stats = "n.risk", hjust = "right") 

I would really appreaciate if you could add the first suggestion, since this is rather annoying. Also it would allow me to set expand to c(0, 0) so the axes would start at the same spot. The other suggestions would be the icing on the cake.

[ddsjoberg]

Thanks for the post @jpnolte !

Find responses below. Please note that the defaults were set to meet many needs. But we know that it does not meet everyone's need, which is why we try to make it possible modify the defaults.

1. You're calling the scale_ggsurvfit() function which reduces the padding/spacing. If you don't make this call, or you modify the defaults padding values, you can make any display fit.
2. Can you pass theme = list(theme_risktable_default(), theme(axis.text.y = element_text(hjust = 0))) to get what you want?
3. I think these options can also be passed to the theme argument, right?
4. Any arguments passed in add_risktable(...) are directly passed to geom_text(...). Will that work for you?

[jpnolte]

Thank you so much for the quick response. You are absolutely right, (2) and (3) can be achieved by providing the arguments through theme() in the list, nice! So this way you can change the position of the risk table title and also the alignment of the axis text. Maybe adding an example with the theme() in the list to the documentation is also helpful to others?

survfit2(Surv(time, status) ~ sex, data = df_lung) |>
  ggsurvfit() +
  scale_x_continuous(expand = c(0.005, 0), 
                     breaks = seq(0, 34, 2),
                     limits = c(0, 34)) + 
  scale_y_continuous(expand = c(0, 0.05), 
                     breaks = seq(0, 1, 0.25),
                     labels = scales::label_percent()) + 
  add_risktable(
    risktable_stats = "n.risk",
    theme = list(
      theme_risktable_default(), 
      theme(
        axis.text.y = element_text(hjust = 0, margin = margin(r = 10)),
        plot.caption = element_text(hjust = 0),
        plot.title.position = "plot"
      )
    )
  ) 

ad 1) Yes, this is exactly my point. I think, and I know others that agree, it looks better the less empty space there is between the actual start of the y and x axis and the first break. In the example above I reduced the empty space causing the overlap. I was thinking that the margin argument can change that, but of course the risk table ends at the same position as the plot, which is the actual problem of the numbers disappearing. I think, this might be a limitation of ggplot? It is not a big problem when you only have 138 patients, since the empty space I have to add to avoid overlapping is quite small. But the bigger the numbers, the more empty space is necessary.

ad 4) I already tried to provide the hjust argument directly to geom_text(), but apparently ggplot will not center and right align them, but instead move the numbers slightly to the left. Than it is of course slightly off in comparison to the axis text of the x axis of the plot.

survfit2(Surv(time, status) ~ sex, data = df_lung) |>
  ggsurvfit() +
  scale_x_continuous(expand = c(0.04, 0), 
                     breaks = seq(0, 34, 2),
                     limits = c(0, 34)) + 
  scale_y_continuous(expand = c(0, 0.05), 
                     breaks = seq(0, 1, 0.25),
                     labels = scales::label_percent()) + 
  add_risktable(
    risktable_stats = "n.risk",
    theme = list(
      theme_risktable_default(), 
      theme(
        axis.text.y = element_text(hjust = 0, margin = margin(r = 5)),
        plot.caption = element_text(hjust = 0),
        plot.title.position = "plot"
      )
    ),
    hjust = 1
  ) 

If you would then also right align the axis text on the x axis you can see how ggplot handles the alignment, moving both away from the ticks.

survfit2(Surv(time, status) ~ sex, data = df_lung) |>
  ggsurvfit() +
  scale_x_continuous(expand = c(0.04, 0), 
                     breaks = seq(0, 34, 2),
                     limits = c(0, 34)) + 
  scale_y_continuous(expand = c(0, 0.05), 
                     breaks = seq(0, 1, 0.25),
                     labels = scales::label_percent()) + 
  add_risktable(
    risktable_stats = "n.risk",
    theme = list(
      theme_risktable_default(), 
      theme(
        axis.text.y = element_text(hjust = 0, margin = margin(r = 5)),
        plot.caption = element_text(hjust = 0),
        plot.title.position = "plot"
      )
    ),
    hjust = 1
  ) + 
  theme(axis.text.x = element_text(hjust = 1))

So I think this is also a limitation of ggplot?

[ddsjoberg]

I think updating an example for the theme argument is a great idea, thanks.

1. I am not sure of a better way to do this. I agree that if you begin with 10,000 subjects at risk, it doesn't look great including the extra padding, but I am not sure of a better way to do it, while also letting user modify the primary plot in any way they'd like while ensuring proper alignment of the risktable with the figure.

2. geom_text() is hard to work with!! We're in agreement there!

[psoldath]

Would it be possible to allow the risk table text to extend beyond the plot, i.e. so it isn't clipped, like a coord_cartesian(clip = "off") for the risk table? In this way, large numbers at either end could simply just extend beyond the plot when there is no padding. The user could then manually adjust the risk table axis.text.y accordingly and set the plot.title.position to "plot".

[psoldath]

@ddsjoberg I think an elegant solution could be to design the risk table like in the attached plots. This design allows for large numbers in the risk table without padding as well as long labels for the strata without an undesirable gap between the plot's y-axis and y-axis title. Would it be possible to align the plot and risk table in that way?

[ddsjoberg]

I don't disagree! I don't know exactly how to make that happen. Essentially, we need to assign different padding to the risktable "figures" (they are just ggplots here), while still ensuring the primary figure lines up with the risktables.

What we're doing internally that during the print of these objects, we inspect the primary plots and make modificatons to the risktable plot(s) to ensure the line up with one another. This allows the users to modify the primary plot in any way they'd like and the risktable still lines up (a great feature!). I don't know how we could ensure the alignment while still giving the users the flexibility they also want.

It's probably possible, but to be up front, I don't have the bandwidth to investigate this at the moment (or the foreseeable future), sadly. It's unfortunate, because this would be a great feature (and there are a bunch of other things I want to improve in the package as well).

[ddsjoberg]

@jpnolte there is a yet-to-be-released change coming to ggplot2 that will require ggsurvfit to go to CRAN again. Would you like to submit a PR with a concise theme example in the gallery?

[psoldath]

Hi again @ddsjoberg. I just wanted to follow up on this whole idea of maybe changing the design of the risk tables in case you are coming out with a new release of ggsurvfit for the upcoming ggplot2 changes? The patchwork free() function makes it easy to combine the plot and risk table without a left shift of the plot's y-axis title from long strata labels. In this way we can easily mimic a design like in the plots above that does not require any padding. Will something like the following be easy to implement in the package? In this code, I create a fake strata to act as a header of the risk table, I think it looks better than using a plot title as the header, and it is still pretty easy to customize further e.g. make it bold etc.

library(tidyverse)
library(survival)
library(ggsurvfit)
library(patchwork)

strata_header_lab <- "At risk (censored)"
x_breaks <- seq(0, 10, 2)
x_max <- 10

survfit <- survfit2(Surv(time, status) ~ surg, data = df_colon) |>
  tidy_survfit()

survplot <- survfit |>
  ggplot(aes(time, estimate, color = strata)) +
  geom_step() +
  scale_x_continuous(breaks = x_breaks) +
  coord_cartesian(
    xlim = c(0, 10),
    ylim = c(0, 1),
    expand = FALSE,
    clip = "off"
  ) +
  theme_classic() +
  theme(legend.position = "none")

risktable <- survfit2(Surv(time, status) ~ surg, data = df_colon) |>
  tidy_survfit(times = x_breaks) |>
  mutate(label = paste0(n.risk, " (", n.censor, ")")) |>
  bind_rows(tibble(time = NA, strata = strata_header_lab)) |>
  mutate(
    strata = factor(
      strata,
      levels = rev(c(
        strata_header_lab,
        sort(setdiff(unique(strata), strata_header_lab))
      ))
    )
  )

risktableplot <- risktable |>
  ggplot(aes(x = time, y = strata, label = label)) +
  geom_text(size = 11 * 0.8 / .pt) +
  scale_x_continuous(breaks = x_breaks) +
  coord_cartesian(expand = FALSE, clip = "off") +
  theme_classic() +
  theme(
    axis.title = element_blank(),
    axis.line = element_blank(),
    axis.ticks = element_blank(),
    axis.text.x = element_blank(),
    axis.text.y.left = element_text(margin = margin(r = 20))
  )

plot <- free(survplot, type = "space", side = "l") /
  risktableplot +
  plot_layout(heights = c(6, 1))

[ddsjoberg]

@psoldath it looks perfect! Thank you for sharing! I didn't know about patchwork:free() !

I think this will be a straightforward drop-in replacement of the existing method. I hope I can get to this in the summer FYI

[ShreyaSreeram27]

i think the above mentioned patchwork:free() should be in a separate pr? @ddsjoberg

[ShreyaSreeram27]

hi @ddsjoberg I'd like to implement the following changes to address this feature request while maintaining backward compatibility and safety:

New Parameters Added:

1. clip = "on" - Controls text clipping (default maintains the current behavior)
2. plot.title.position = "panel" - Better title positioning control
3. ... - Accepts additional ggplot2 theme elements for full customization
   I'll also add clear documentation with examples showcasing how to use this.

the patchwork:free solution seems more complex and so i will get to it in a sep pr (will also have to think how/if it would complement the above clip solution or would be redundant to have)

[ShreyaSreeram27]

@ddsjoberg seems like most of ggsurvfit_align_plots can be restructured to way fewer lines of code, thinking something like

combine_plots_free <- function(main_plot, risk_table) {
patchwork::free(main_plot, type = "space", side = "l") /
risk_table +
patchwork::plot_layout(heights = c(6, 1))
}

replacing the title with the fake strata approach would require some more change.. are we looking to do that as well?

[ddsjoberg]

@ShreyaSreeram27 , awesome! Yeah I think we'd need to update this whole pipeline. Can you outline the updates needed and we can meet next week to review together?