Definition of a gtsummary Object

This vignette is meant for those who wish to contribute to {gtsummary}, or users who wish to gain an understanding of the inner-workings of a {gtsummary} object so they may more easily modify them to suit your own needs. If this does not describe you, please refer to the {gtsummary} website to an introduction on how to use the package’s functions and tutorials on advanced use.

Introduction

Every {gtsummary} object has a few characteristics common among all objects. Here, we review those characteristics, and provide instructions on how to construct a {gtsummary} object.

library(gtsummary)
library(purrr); library(dplyr); library(tibble)

tbl_regression_ex <-
  lm(age ~ grade + marker, trial) %>%
  tbl_regression() %>%
  bold_p(t = 0.5) 

tbl_summary_ex <-
  trial %>%
  select(trt, age, grade, response) %>%
  tbl_summary(by = trt)

Structure of a {gtsummary} object

Every {gtsummary} object is a list comprising of, at minimum, these elements:

.$table_body    .$table_header         

table_body

The .$table_body object is the data frame that will ultimately be printed as the output. The table must include columns "label", "row_type", and "variable". The "label" column is printed, and the other two are hidden from the final output.

tbl_summary_ex$table_body
variable row_type label stat_1 stat_2
age label Age, yrs 46 (37, 59) 48 (39, 56)
age missing Unknown 7 4
grade label Grade
grade level I 35 (36%) 33 (32%)
grade level II 32 (33%) 36 (35%)
grade level III 31 (32%) 33 (32%)
response label Tumor Response 28 (29%) 33 (34%)
response missing Unknown 3 4

table_header

The .$table_header object is a data frame containing information about each of the columns in .$table_body (one row per column in .$table_body). The table header has the following columns:

Column Description
column Column name from table_body
label Label that will be displayed (if column is displayed in output)
hide Logical indicating whether the column is hidden in the output
align Specifies the alignment/justification of the column, e.g. ‘center’ or ‘left’
missing_emdash Indicates the rows to include an em-dash for missing cells. For example row_ref == TRUE in tbl_regression()
indent String of R code that results in a logical vector that specifies rows to indent, e.g. row_type != 'label'
text_interpret the {gt} function that is used to interpret the column label
bold For columns that bold rows conditionally, the column includes a string of R code that results in a logical vector indicating the rows to bold For example, row_type == 'label'
italic For columns that italicize rows conditionally, the column includes a string of R code that results in a logical vector indicating the rows to italicize. For example, row_type == 'label'
fmt_fun If the column needs to be formatted, this list column contains the function that performs the formatting. Note, this is the function object; not the character name of a function.
footnote_abbrev Lists the abbreviation footnotes for a table. All abbreviation footnotes are collated into a single footnote. For example, ‘OR = Odds Ratio’ and ‘CI = Confidence Interval’ appear in a single footnote.
footnote Lists the footnotes that will appear for each column.
spanning_header Includes text printed above columns as spanning headers. See tbl_merge(...)$table_header output for example of use.

NOTE: Columns ‘hide’, ‘align’, ‘missing_emdash’, ‘indent’, ‘bold’, and ‘italic’ MUST follow the tidyverse style guidelines and include spaces around any variable names, e.g. row_type == 'label' (NOT row_type=='label').

Example from tbl_regression()

tbl_regression_ex$table_header
column label hide align missing_emdash indent text_interpret bold italic fmt_fun footnote_abbrev footnote spanning_header
variable variable TRUE center gt::md NULL
var_type var_type TRUE center gt::md NULL
row_ref row_ref TRUE center gt::md NULL
row_type row_type TRUE center gt::md NULL
label **Characteristic** FALSE left row_type != 'label' gt::md NULL
N N TRUE center gt::md function (x) ifelse(is.na(x), NA, sprintf("%.0f", x))
estimate **Beta** FALSE center row_ref == TRUE gt::md function (x, digits = 2) { map(digits:1, ~glue("abs(x) < 10^{digits - .x} ~ sprintf('%.{.x}f', x)")) %>% paste(collapse = ", ") %>% { glue("case_when({.}, TRUE ~ ifelse(is.na(x), NA_character_, sprintf('%.0f', x)))") } %>% parse(text = .) %>% eval() %>% { ifelse(as.numeric(.) == 0 & str_starts(., pattern = "-"), str_remove(., pattern = "-"), .) } }
conf.low conf.low TRUE center gt::md function (x, digits = 2) { map(digits:1, ~glue("abs(x) < 10^{digits - .x} ~ sprintf('%.{.x}f', x)")) %>% paste(collapse = ", ") %>% { glue("case_when({.}, TRUE ~ ifelse(is.na(x), NA_character_, sprintf('%.0f', x)))") } %>% parse(text = .) %>% eval() %>% { ifelse(as.numeric(.) == 0 & str_starts(., pattern = "-"), str_remove(., pattern = "-"), .) } }
conf.high conf.high TRUE center gt::md function (x, digits = 2) { map(digits:1, ~glue("abs(x) < 10^{digits - .x} ~ sprintf('%.{.x}f', x)")) %>% paste(collapse = ", ") %>% { glue("case_when({.}, TRUE ~ ifelse(is.na(x), NA_character_, sprintf('%.0f', x)))") } %>% parse(text = .) %>% eval() %>% { ifelse(as.numeric(.) == 0 & str_starts(., pattern = "-"), str_remove(., pattern = "-"), .) } }
ci **95% CI** FALSE center row_ref == TRUE gt::md NULL CI = Confidence Interval
p.value **p-value** FALSE center gt::md p.value <= 0.5 function (x, digits = 1, prepend_p = FALSE) { if (digits == 2) { p_fmt <- case_when(x > 1 ~ NA, x < 0 ~ NA, x > 0.99 ~ ">0.99", round(x, 2) >= 0.1 ~ sprintf("%.2f", x), x >= 0.001 ~ sprintf("%.3f", x), x < 0.001 ~ "<0.001") } else if (digits == 1) { p_fmt <- case_when(x > 1 ~ NA, x < 0 ~ NA, x > 0.9 ~ ">0.9", round(x, 1) >= 0.2 ~ sprintf("%.1f", x), round(x, 2) >= 0.1 ~ sprintf("%.2f", x), x >= 0.001 ~ sprintf("%.3f", x), x < 0.001 ~ "<0.001") } else { stop("'digits' argument must be 1 or 2.") } if (prepend_p == TRUE) { p_fmt <- case_when(is.na(p_fmt) ~ NA, stringr::str_sub(p_fmt, end = 1) %in% c("<", ">") ~ paste0("p", p_fmt), TRUE ~ paste0("p=", p_fmt)) } return(p_fmt) }

Constructing a {gtsummary} object

table_body

When constructing a {gtsummary} object, the author will begin with the .$table_body object. Recall the .$table_body data frame must include columns "label", "row_type", and "variable". Of these columns, only the "label" column will be printed with the final results. The "row_type" column typically will control whether or not the label column is indented. The "variable" column is often used in the inline_text() family of functions, and merging {gtsummary} tables with tbl_merge().

tbl_regression_ex %>%
  pluck("table_body") %>%
  select(variable, row_type, label)
variable row_type label
grade label Grade
grade level I
grade level II
grade level III
marker label Marker Level, ng/mL

The other columns in .$table_body are created by the user and are likely printed in the output. Formatting and printing instructions for these columns is stored in .$table_header.

table_header

The .$table_header has one row for every column in .$table_body containing instructions how to format each column, the column headers, and more. There are a few internal {gtsummary} functions to assist in constructing and modifying a .$table_header data frame.

First is the table_header_fill_missing() function. This function ensures .$table_header contains a row for every column of .$table_body. If a column does not exist, it is populated with appropriate default values.

gtsummary:::table_header_fill_missing(
  table_header = tibble(column = names(tbl_regression_ex$table_body))
) 
column label hide align missing_emdash indent text_interpret bold italic fmt_fun footnote_abbrev footnote spanning_header
variable variable TRUE center gt::md NULL
var_type var_type TRUE center gt::md NULL
row_ref row_ref TRUE center gt::md NULL
row_type row_type TRUE center gt::md NULL
label label TRUE left row_type != 'label' gt::md NULL
N N TRUE center gt::md NULL
estimate estimate TRUE center gt::md NULL
conf.low conf.low TRUE center gt::md NULL
conf.high conf.high TRUE center gt::md NULL
ci ci TRUE center gt::md NULL
p.value p.value TRUE center gt::md NULL

The modify_header_internal() is useful for assigning column headers. The function accepts a complete {gtsummary} object as its input, and returns an updated version where the column labels have been added to .$table_header. The function also switches the default .$table_header$hide from TRUE to FALSE, resulting in column with labels being printed.

Printing a {gtsummary} object

All {gtsummary} objects are printed with print.gtsummary(). But before a {gtsummary} object is printed, it is converted to a {gt} object using as_gt(). This function takes the {gtsummary} object as its sole input, and uses the information in .$table_header to construct a list of {gt} calls that will be executed on .$table_body. After the {gtsummary} object is converted to {gt}, it is then printed as any other {gt} object.

In some cases, the package defaults to printing with knitr::kable() utilizing the as_kable() function.

While the actual print function is slightly more involved, it is basically this:

print.gtsummary <- function(x) {
  if (getOption("gtsummary.print_engine") == "gt") {
    return(as_gt(x) %>% print())
  }
  else if (getOption("gtsummary.print_engine") == "kable") {
    return(as_kable(x) %>% print())
  }
}