Below is an example, but we will need to rename the function to align with the other CI functions. We will also need to generalize this a bit to account for different boot.ci(type) arguments and perhaps even boot(stype) arguments.

The other CI functions also include N in the results, so we should do the same with this one.

Also account for the special case with N = 1, when boot::boot.ci() doesn't error but doesn't return results: we want to the same structure of the returned named list each time.

boot_mean_ci <- function(x, conf.level = 0.95, R = 2000) {
  result <- list(estimate = NULL, conf.low = NULL, conf.high = NULL, conf.level = conf.level, R = R)
  boot <- boot::boot(data = x, statistic = \(x, i) mean(x[i], na.rm = TRUE), R = R, stype = "i")
  boot.ci <- boot::boot.ci(boot, conf = conf.level, type = "perc")
  result$estimate <- boot$t0
  result$conf.low <- boot.ci$percent[1, 4] |> unname()
  result$conf.high <- boot.ci$percent[1, 5] |> unname()

  result
}

cards::ard_continuous(ADSL, variables = AGE, statistic = everything() ~ boot_mean_ci)

I am torn on whether we should add this as an option to ard_continuous_ci(), because we would need to expose the type and stype arguments, if we do this then where would we stop? For now, let's not add...any other thoughts?