Feature request: Add type = "cloglog" (log minus log survival)

[avishaitsur]

The problem: Log minus log survival curves are widely used to assess the assumption of proportionality of hazards. While plotting these curves is available using the survival and survminer packages, it is currently unavailable in ggsurvfit.

Proposed solution: Add an option of type = "cloglog" to ggsurvfit::tidy_survfit. Specifically, this would need to be added in the:

• Function call

  tidy_survfit <- function(x,
                       times = NULL,
                       type = c("survival", "risk", "cumhaz", "cloglog"))

• Select transformation function

  if (rlang::is_string(type)) {
  .transfun <-
    switch(type,
           survival = function(y) y,
           cuminc = function(y) y,
           risk = function(y) 1 - y,
           # survfit object contains an estimate for Cumhaz and SE based on Nelson-Aalen with or without correction for ties
           # However, no CI is calculated automatically. For plotting, the MLE estimator is used for convenience.
           cumhaz = function(y) -log(y),
           cloglog = function(y) log(-log(y))
    )
  } else {
  .transfun <- type
  }
  if (!rlang::is_function(.transfun)) {
  cli_abort("The {.var type} argument must be one of {.val {c('survival', 'risk', 'cumhaz', 'cloglog')}}, or a function.")
  }

• Add estimate type label:

    estimate_type_label =
      dplyr::case_when(
        rlang::is_string(.env$type) && .env$type %in% "survival" ~ "Survival Probability",
        rlang::is_string(.env$type) && .env$type %in% "cuminc" ~ "Cumulative Incidence",
        rlang::is_string(.env$type) && .env$type %in% "risk" ~ "Risk",
        rlang::is_string(.env$type) && .env$type %in% "cumhaz" ~ "Cumulative Hazard",
        rlang::is_string(.env$type) && .env$type %in% "cloglog" ~ "Log Minus Log Survival",
        TRUE ~ rlang::expr_deparse(.transfun)
      )

• Monotonicity

  x$monotonicity_type <-
    switch(type,
           "survival" = "decreasing",
           "cuminc" = "increasing",
           "risk" = "increasing",
           "cumhaz" = "increasing",
           "cloglog" = "increasing"
    )

• Initial survival estimates of value one should be ignored as log(-log(1)) reaches infinity.
• Unlike the other figures, the x-axis is commonly on the log scale.

Example of a figure using the survival package

Example of the desired figure using the ggsurvfit package

[ddsjoberg]

Dear @avishaitsur ,

i think that is a great request. I see you've done the research into he package's internals. Would you like to submit a PR to add this?