[Experimental] The tbl_custom_summary() function calculates descriptive statistics for continuous, categorical, and dichotomous variables. This function is similar to tbl_summary() but allows you to provide a custom function in charge of computing the statistics (see Details).

tbl_custom_summary(
  data,
  by = NULL,
  label = NULL,
  stat_fns,
  statistic,
  digits = NULL,
  type = NULL,
  value = NULL,
  missing = NULL,
  missing_text = NULL,
  include = everything(),
  overall_row = FALSE,
  overall_row_last = FALSE,
  overall_row_label = NULL
)

Arguments

data

A data frame

by

A column name (quoted or unquoted) in data. Summary statistics will be calculated separately for each level of the by variable (e.g. by = trt). If NULL, summary statistics are calculated using all observations. To stratify a table by two or more variables, use tbl_strata()

label

List of formulas specifying variables labels, e.g. list(age ~ "Age", stage ~ "Path T Stage"). If a variable's label is not specified here, the label attribute (attr(data$age, "label")) is used. If attribute label is NULL, the variable name will be used.

stat_fns

Formula or list of formulas specifying the function to be used to compute the statistics (see below for details and examples). You can also use dedicated helpers such as continuous_summary(), ratio_summary() or proportion_summary().

statistic

List of formulas specifying the glue::glue() pattern to display the statistics for each variable. The statistics should be returned by the functions specified in stat_fns (see below for details and examples).

digits

List of formulas specifying the number of decimal places to round summary statistics. If not specified, tbl_summary guesses an appropriate number of decimals to round statistics. When multiple statistics are displayed for a single variable, supply a vector rather than an integer. For example, if the statistic being calculated is "{mean} ({sd})" and you want the mean rounded to 1 decimal place, and the SD to 2 use digits = list(age ~ c(1, 2)). User may also pass a styling function: digits = age ~ style_sigfig

type

List of formulas specifying variable types. Accepted values are c("continuous", "continuous2", "categorical", "dichotomous"), e.g. type = list(age ~ "continuous", female ~ "dichotomous"). If type not specified for a variable, the function will default to an appropriate summary type. See below for details.

value

List of formulas specifying the value to display for dichotomous variables. gtsummary selectors, e.g. all_dichotomous(), cannot be used with this argument. See below for details.

missing

Indicates whether to include counts of NA values in the table. Allowed values are "no" (never display NA values), "ifany" (only display if any NA values), and "always" (includes NA count row for all variables). Default is "ifany".

missing_text

String to display for count of missing observations. Default is "Unknown".

include

variables to include in the summary table. Default is everything()

overall_row

Logical indicator to display an overall row. Default is FALSE. Use add_overall() to add an overall column.

overall_row_last

Logical indicator to display overall row last in table. Default is FALSE, which will display overall row first.

overall_row_label

String indicating the overall row label. Default is "Overall".

Value

A tbl_custom_summary and tbl_summary object

Similarities with tbl_summary()

Please refer to the help file of tbl_summary() regarding the use of select helpers, and arguments include, by, type, value, digits, missing and missing_text.

stat_fns argument

The stat_fns argument specify the custom function(s) to be used for computing the summary statistics. For example, stat_fns = everything() ~ foo.

Each function may take the following arguments: foo(data, full_data, variable, by, type, ...)

  • data= is the input data frame passed to tbl_custom_summary(), subset according to the level of by or variable if any, excluding NA values of the current variable

  • full_data= is the full input data frame passed to tbl_custom_summary()

  • variable= is a string indicating the variable to perform the calculation on

  • by= is a string indicating the by variable from tbl_custom_summary=, if present

  • type= is a string indicating the type of variable (continuous, categorical, ...)

  • stat_display= a string indicating the statistic to display (for the statistic argument, for that variable)

The user-defined does not need to utilize each of these inputs. It's encouraged the user-defined function accept ... as each of the arguments will be passed to the function, even if not all inputs are utilized by the user's function, e.g. foo(data, ...) (see examples).

The user-defined function should return a one row dplyr::tibble() with one column per summary statistics (see examples).

statistic argument

The statistic argument specifies the statistics presented in the table. The input is a list of formulas that specify the statistics to report. For example, statistic = list(age ~ "{mean} ({sd})"). A statistic name that appears between curly brackets will be replaced with the numeric statistic (see glue::glue()). All the statistics indicated in the statistic argument should be returned by the functions defined in the stat_fns argument.

When the summary type is "continuous2", pass a vector of statistics. Each element of the vector will result in a separate row in the summary table.

For both categorical and continuous variables, statistics on the number of missing and non-missing observations and their proportions are also available to display.

  • {N_obs} total number of observations

  • {N_miss} number of missing observations

  • {N_nonmiss} number of non-missing observations

  • {p_miss} percentage of observations missing

  • {p_nonmiss} percentage of observations not missing

Note that for categorical variables, {N_obs}, {N_miss} and {N_nonmiss} refer to the total number, number missing and number non missing observations in the denominator, not at each level of the categorical variable.

It is recommended to use modify_footnote() to properly describe the displayed statistics (see examples).

Caution

The returned table is compatible with all gtsummary features applicable to a tbl_summary object, like add_overall(), modify_footnote() or bold_labels().

However, some of them could be inappropriate in such case. In particular, add_p() do not take into account the type of displayed statistics and always return the p-value of a comparison test of the current variable according to the by groups, which may be incorrect if the displayed statistics refer to a third variable.

Example Output

Example 1

Example 2

Example 3

Author

Joseph Larmarange

Examples

# \donttest{
# Example 1 ----------------------------------
my_stats <- function(data, ...) {
  marker_sum = sum(data$marker, na.rm = TRUE)
  mean_age = mean(data$age, na.rm = TRUE)
  dplyr::tibble(
    marker_sum = marker_sum,
    mean_age = mean_age
  )
}

my_stats(trial)
#> # A tibble: 1 × 2
#>   marker_sum mean_age
#>        <dbl>    <dbl>
#> 1       174.     47.2

tbl_custom_summary_ex1 <-
  trial %>%
  tbl_custom_summary(
    include = c("stage", "grade"),
    by = "trt",
    stat_fns = everything() ~ my_stats,
    statistic = everything() ~ "A: {mean_age} - S: {marker_sum}",
    digits = everything() ~ c(1, 0),
    overall_row = TRUE,
    overall_row_label = "All stages & grades"
  ) %>%
  add_overall(last = TRUE) %>%
  modify_footnote(
    update = all_stat_cols() ~ "A: mean age - S: sum of marker"
  ) %>%
  bold_labels()

# Example 2 ----------------------------------
# Use `data[[variable]]` to access the current variable
mean_ci <- function(data, variable, ...) {
  test <- t.test(data[[variable]])
  dplyr::tibble(
    mean = test$estimate,
    conf.low = test$conf.int[1],
    conf.high = test$conf.int[2]
  )
}

tbl_custom_summary_ex2 <-
  trial %>%
  tbl_custom_summary(
    include = c("marker", "ttdeath"),
    by = "trt",
    stat_fns = ~ mean_ci,
    statistic = ~ "{mean} [{conf.low}; {conf.high}]"
  ) %>%
  add_overall(last = TRUE) %>%
  modify_footnote(
    update = all_stat_cols() ~ "mean [95% CI]"
  )

# Example 3 ----------------------------------
# Use `full_data` to access the full datasets
# Returned statistic can also be a character
diff_to_great_mean <- function(data, full_data, ...) {
  mean <- mean(data$marker, na.rm = TRUE)
  great_mean <- mean(full_data$marker, na.rm = TRUE)
  diff <- mean - great_mean
  dplyr::tibble(
    mean = mean,
    great_mean = great_mean,
    diff = diff,
    level = ifelse(diff > 0, "high", "low")
  )
}

tbl_custom_summary_ex3 <-
  trial %>%
  tbl_custom_summary(
    include = c("grade", "stage"),
    by = "trt",
    stat_fns = ~ diff_to_great_mean,
    statistic = ~ "{mean} ({level}, diff: {diff})",
    overall_row = TRUE
  ) %>%
  bold_labels()
# }