Tables in Clinical Trials with R

3.1.1 gt

The gt package lets you add a title and even a subtitle and preheader lines (for RTF) with its tab_header() function. In the following example, we create some sample_data and feed that into the gt() function. We can automatically create a table stub (for row labels) and row groups with the rowname_col and groupname_col arguments of gt().

resetSession()
library(gt)

sample_data <-
  dplyr::tibble(
    label = c("n", "Mean (SD)", "Median", "Min - Max", "F", "M", "U", "UNDIFFERENTIATED"),
    `val_A: Drug X` = c(134, 33.8, 33, NA, 79, 51, 3, 1),
    `val_B: Placebo` = c(134, 35.4, 35, NA, 77, 55, 2, 0),
    `val_C: Combination` = c(132, 35.4, 35, NA, 66, 60, 4, 2),
    category = c(rep("Age (Years)", 4), rep("Sex, n (%)", 4))
  )

gt_tbl <-
  gt(
    sample_data, 
    rowname_col = "label",
    groupname_col = "category"
  ) |>
  tab_header(
    title = "x.x: Study Subject Data",
    subtitle = md(
      "x.x.x: Demographic Characteristics  \n Table x.x.x.x: Demographic
      Characteristics - Full Analysis Set"
    ),
    preheader = c("Protocol: XXXXX", "Cutoff date: DDMMYYYY")
  ) |> 
  tab_source_note("Source: ADSL DDMMYYYY hh:mm; Listing x.xx; SDTM package: DDMMYYYY") |> 
  sub_missing(missing_text = "") |> 
  tab_options(
    page.orientation = "landscape",
    page.numbering = TRUE,
    page.header.use_tbl_headings = TRUE,
    page.footer.use_tbl_notes = TRUE
  )

gt_tbl

The above example contains the use of the tab_source_note() function. You can create as many source notes in the table footer as you need, and they typically describe the data table as a whole (i.e., not pointing to anything specific). For that, you can use footnotes and target cells that require additional explanation. Here’s an example of that using tab_footnote():

gt_tbl |>
  tab_footnote(
    footnote = "This is the combination of the two.",
    locations = cells_column_labels(columns = `val_C: Combination`)
  ) |>
  tab_footnote(
    footnote = "These values are the same.",
    locations = cells_body(
      columns = matches("_A|_B"), rows = "n"
      )
  )

The tab_footnote() function allows for footnotes to be placed anywhere in the table (using the cells_*() helper functions for targeting). Targeting columns, rows, or other locations can be done with Tidyselect-style helper functions (e.g., matches(), starts_with(), etc.), ID values, or indices.

As a final note on the first example, we can specify certain page.* options that make RTF output ideal for regulatory filing purposes. The options employed above in the tab_options() call ensure that pages are in landscape orientation, page numbering for each table is activated, and that page header and footer are used for the table’s headings and footer elements.

3.1.2 rtables

The basic_table() function in rtables has the arguments titles, subtitles, main_footer, prov_footer to add titles and footnotes to tables. rtables also supports referential footnotes.

So for example a basic demographics table created with rtables via tern with title and footnotes would look as follows:

resetSession()
library(rtables)

lyt <- basic_table(
  title = "Demographic Table - All Patients",
  subtitles = c("Cutoff Date: June 01, 2022", "Arm B received a placebo."),
  main_footer = c("Missing data is omitted.")
) |>
  split_cols_by("ARM") |>
  analyze(c("AGE", "SEX"))

build_table(lyt, adsl)

Demographic Table - All Patients
Cutoff Date: June 01, 2022
Arm B received a placebo.

————————————————————————————————————————————————
         A: Drug X   B: Placebo   C: Combination
————————————————————————————————————————————————
AGE                                             
  Mean     33.77       35.43          35.43     
SEX                                             
  F         79           82             70      
  M         55           52             62      
————————————————————————————————————————————————

Missing data is omitted.

3.1.3 flextable

Titles and notes can be added and formatted with the flextable package. It is possible to add them in the header and in the footer. Several methods are possible but for most needs, the add_header_lines() and add_footer_lines() functions will be the easiest to use.

Let’s create first a flextable from an aggregation that will be used to illustrate the features.

resetSession()
library(flextable)
library(dplyr)

z <- adsl |> 
  group_by(ARM, SEX) |>
  summarise(avg = mean(AGE), sd = sd(AGE)) |>
  tabulator(rows = "SEX", columns = "ARM",
    Y = as_paragraph(avg, " (", sd, ")")) |> 
  as_flextable()
z

The following shows how to add titles or notes:

z |>
  add_header_lines("hello world") |>
  add_footer_lines("bye bye world")

For Word output, users can prepend a table number that will auto-incremente.

docx_file <- "reports/flextable-title-01.docx"
ft <- add_header_lines(z, "hello world") |>
  prepend_chunks(
    i = 1, j = 1, part = "header", 
    as_chunk("Table "), as_word_field("SEQ tab \u005C* Arabic"), 
    as_chunk(": ")) |>
  add_footer_lines("bye bye world") |>
  theme_vanilla()
save_as_docx(ft, ft, path = docx_file)

Footnotes are also available in flextable with function footnote(). The function lets users add footnotes and references to it on the table.

footnote(z, i = c(1, 2, 2), j = c(1, 5, 7), 
         value = as_paragraph("hello world"), ref_symbols = "(1)")

3.1.4 tfrmt

The tfrmt() function in the tfrmt package includes the arguments title and subtitle to add titles. Within the footnote_plan() function, the user can nest multiple footnote_structures to add footnotes with superscript reference symbols on groups, columns or labels.

To demonstrate, this example will create a mock demographics table:

resetSession()
library(tfrmt)
library(dplyr)
library(tidyr)

# Create mock data
df <- crossing(group = c("AGE", "SEX"),
               label = c("label 1", "label 2"),
               column = c("Drug X", "Placebo", "Combination"),
               param = c("count", "percent"))

# Create specification
tfrmt_spec <- tfrmt(
  # Add titles
  title = "Demographic Table - All Patients",
  subtitle = "Cutoff Date: June 01, 2022. Arm B received a placebo.",

  # Specify table features
  group = group,
  label = label,
  column = column,
  param = param,
  row_grp_plan = row_grp_plan(
    row_grp_structure(group_val = ".default",
                      element_block(post_space = "   ")) ),

  # Define cell formatting
  body_plan = body_plan(
    frmt_structure(group_val = ".default", label_val = ".default",
                   frmt_combine("{count} ({percent})",
                                count = frmt("xx"),
                                percent = frmt("xx.x")))),

  # Add footnotes here
  footnote_plan = footnote_plan(
    footnote_structure(footnote_text = "Footnote about column", column_val = "Combination"),
    footnote_structure(footnote_text = "Footnote about group", group_val = "AGE"),
    marks = "numbers"),

)

print_mock_gt(tfrmt_spec, df)

See this vignette for more details on footnotes: link to website

3.1.5 tables

The tables package concentrates on the table itself. The titles are generally written as part of the surrounding document. Footnotes would be added after constructing the table by modifying individual entries.

Alternatively for HTML output, only the footnote markers need to be added by modifying entries, and then the footnotes can be applied by using toHTML(tab, options = list(doFooter = TRUE, HTMLfooter = HTMLfootnotes(...)).

resetSession()
adsl <- cadsl

library(tables)
table_options(doCSS = TRUE)

sd_in_parens <- function(x) sprintf("(%.1f)", sd(x))

tab <- tabular(SEX ~ Heading()*ARM*
                     Heading()*AGE*
                     Heading()*(mean + sd_in_parens), 
               data = adsl)

rowLabels(tab)[1,1] <- paste(rowLabels(tab)[1,1], "<sup>a</sup>")
tab[2,2] <- sprintf("%s%s", tab[2,2], "<sup>b</sup>")
tab[2,3] <- sprintf("%.2f%s", tab[2,3], "<sup>b</sup>")

footnotes <- HTMLfootnotes(tab, a = "This is a label footnote.",
                                b = "These are cell footnotes.")
toHTML(tab, options = list(HTMLfooter = footnotes,
                           doFooter = TRUE))

3.1.6 tidytlg

The gentlg() function in the tidytlg package includes the title argument for adding title and the footers argument for adding footnotes to the table output. Users can include a vector of character strings for multiple lines of footnotes (please see an example below). At the bottom line of the footnotes, the file name of the table and the path of the table program along with the datetime stamp are automatically created.

resetSession()
library(dplyr)
library(tidytlg)

adsl <- formatters::ex_adsl 

# create analysis set row
t1 <- freq(adsl,
           rowvar = "ITTFL",
           colvar = "ARM",
           statlist = statlist("n"),
           subset = ITTFL == "Y",
           rowtext = "Analysis set: ITT") 

# create univariate stats for age
t2 <- univar(adsl,
               rowvar = "AGE",
               colvar = "ARM",
               statlist = statlist(c("N", "MEANSD", "MEDIAN", "RANGE", "IQRANGE")),
               row_header = "Age (years)",
               decimal = 0)

tbl <- bind_table(t1, t2)

# assign table id
tblid <- "Table01"

# output the analysis results
gentlg(huxme       = tbl,
       format      = "HTML",
       print.hux = FALSE,
       file        = tblid,
       orientation = "portrait",
       title = "Demographic and Baseline Characteristics; Intent-to-treat Analysis Set",
       footers = c("Key: IQ = Interquartile","Note: N reflects non-missing values"),
       colheader = c("","A: Drug X","B: Placebo","C: Combination"))

To programmatically incorporate titles and footnotes into each table program, users can create an excel file called titles.xls (see below snapshot) with the columns of "TABLE ID","IDENTIFIER","TEXT". In the gentlg() function call, users just need to provide the argument of title_file for specifying the location of titles.xls. Then the title and footnotes corresponding to the table ID will be automatically included in the table output. Users need to make sure the correct table ID is used for the file argument of the gentlg() function call.

gentlg(huxme       = tbl,
       format      = "HTML",
       print.hux = FALSE,
       file        = tblid,
       orientation = "portrait",
       title_file = system.file("extdata/titles.xls", package = "tidytlg"),
       colheader = c("","A: Drug X","B: Placebo","C: Combination"))

3.2.1 flextable

The set_caption() function in flextable is the recommended way to add captions.

resetSession()
library(flextable)

flextable(head(cars)) |>
  set_caption(
    caption = "a caption",
    autonum = officer::run_autonum(seq_id = "tab", bkm = "flextable-label"))

In bookdown, use the syntax \@ref(tab:flextable-label) to create a linked reference to the table. Here is an example of a reference: 3.2.

With Quarto, the R chunk code should be transformed as:

#| label: tbl-flextable-label
#| tbl-cap: a caption
flextable(head(cars))

3.2.2 tables

As with titles, captions would be added as part of the surrounding document rather than part of the table object.

3.3.1 rtables

rtables supports context-preserving pagination in both the horizontal and vertical directions (via the interface provided by formatters) via calling paginate_table() directly, and within the export_as_* rendering functions. Users specify page dimensions (in either inches, or lines long and characters wide) and font information and the pagination and export machinery paginates the table such that each portion will fit fully on its page when rendered as text, including title, footer, and referential footnote materials.

For vertical pagination, summary rows (whether label rows, or so-called content rows containing summary values) are repeated after page breaks to preserve context on the following page. We see this in action below when pagination occurs within the strtata B – ASIAN facet of the the following table

resetSession()
library(rtables)

lyt <- basic_table(title = "main title", subtitles = "subtitle", main_footer = "main footer", prov_footer = "provenance footer") |>
  split_cols_by("ARM") |>
  split_cols_by("SEX", split_fun = keep_split_levels(c("F", "M"))) |>
  split_rows_by("STRATA1", split_fun = keep_split_levels(c("A", "B"))) |>
  split_rows_by("RACE", split_fun = keep_split_levels(c("ASIAN", "WHITE"))) |>
  summarize_row_groups() |>
  analyze("AGE", afun = function(x, ...) in_rows("mean (sd)" = rcell(c(mean(x), sd(x)), format = "xx.x (xx.x)"),
                                                 "range" = rcell(range(x), format = "xx.x - xx.x")))

tbl <- build_table(lyt, ex_adsl)
tbl

main title
subtitle

—————————————————————————————————————————————————————————————————————————————————————————————————
                        A: Drug X                  B: Placebo                C: Combination      
                     F             M             F             M             F             M     
—————————————————————————————————————————————————————————————————————————————————————————————————
A                                                                                                
  ASIAN         11 (13.9%)    10 (19.6%)    14 (18.2%)    10 (18.2%)    11 (16.7%)     7 (11.7%) 
    mean (sd)   29.0 (3.9)    35.0 (6.1)    31.1 (5.5)    40.9 (10.3)   33.7 (4.0)    37.0 (5.9) 
    range       24.0 - 35.0   28.0 - 43.0   23.0 - 46.0   27.0 - 62.0   28.0 - 40.0   28.0 - 47.0
  WHITE          5 (6.3%)      3 (5.9%)      3 (3.9%)      3 (5.5%)      3 (4.5%)      5 (8.3%)  
    mean (sd)   34.4 (2.9)    35.3 (8.5)    33.7 (2.9)    38.7 (10.3)   29.7 (6.1)    32.2 (7.7) 
    range       30.0 - 37.0   29.0 - 45.0   32.0 - 37.0   30.0 - 50.0   23.0 - 35.0   25.0 - 45.0
B                                                                                                
  ASIAN         11 (13.9%)     9 (17.6%)    15 (19.5%)     7 (12.7%)    11 (16.7%)    14 (23.3%) 
    mean (sd)   29.5 (5.7)    35.3 (7.1)    38.7 (10.0)   37.6 (10.6)   41.5 (9.6)    36.1 (7.5) 
    range       23.0 - 40.0   27.0 - 48.0   26.0 - 58.0   26.0 - 58.0   32.0 - 64.0   25.0 - 48.0
  WHITE          5 (6.3%)      4 (7.8%)      8 (10.4%)     4 (7.3%)      5 (7.6%)      2 (3.3%)  
    mean (sd)   35.0 (3.4)    39.0 (11.2)   32.2 (5.3)    33.0 (9.8)    33.4 (6.5)    29.0 (4.2) 
    range       31.0 - 39.0   24.0 - 48.0   26.0 - 42.0   21.0 - 42.0   28.0 - 44.0   26.0 - 32.0
—————————————————————————————————————————————————————————————————————————————————————————————————

main footer

provenance footer

paginate_table(), then, breaks our table into subtables – including repeated context where appropriate – which will fit on physical pages (we use 5.2 x 3.5 inch “pages” for illustrative purposes here):

paginate_table(tbl, pg_width = 5.2, pg_height = 3.5, min_siblings = 0)

[[1]]
main title
subtitle

———————————————————————————————————————————————————————
                        A: Drug X           B: Placebo 
                     F             M             F     
———————————————————————————————————————————————————————
A                                                      
  ASIAN         11 (13.9%)    10 (19.6%)    14 (18.2%) 
    mean (sd)   29.0 (3.9)    35.0 (6.1)    31.1 (5.5) 
    range       24.0 - 35.0   28.0 - 43.0   23.0 - 46.0
  WHITE          5 (6.3%)      3 (5.9%)      3 (3.9%)  
    mean (sd)   34.4 (2.9)    35.3 (8.5)    33.7 (2.9) 
    range       30.0 - 37.0   29.0 - 45.0   32.0 - 37.0
B                                                      
  ASIAN         11 (13.9%)     9 (17.6%)    15 (19.5%) 
    mean (sd)   29.5 (5.7)    35.3 (7.1)    38.7 (10.0)
———————————————————————————————————————————————————————

main footer

provenance footer

[[2]]
main title
subtitle

———————————————————————————————————————————————————————
                B: Placebo         C: Combination      
                     M             F             M     
———————————————————————————————————————————————————————
A                                                      
  ASIAN         10 (18.2%)    11 (16.7%)     7 (11.7%) 
    mean (sd)   40.9 (10.3)   33.7 (4.0)    37.0 (5.9) 
    range       27.0 - 62.0   28.0 - 40.0   28.0 - 47.0
  WHITE          3 (5.5%)      3 (4.5%)      5 (8.3%)  
    mean (sd)   38.7 (10.3)   29.7 (6.1)    32.2 (7.7) 
    range       30.0 - 50.0   23.0 - 35.0   25.0 - 45.0
B                                                      
  ASIAN          7 (12.7%)    11 (16.7%)    14 (23.3%) 
    mean (sd)   37.6 (10.6)   41.5 (9.6)    36.1 (7.5) 
———————————————————————————————————————————————————————

main footer

provenance footer

[[3]]
main title
subtitle

———————————————————————————————————————————————————————
                        A: Drug X           B: Placebo 
                     F             M             F     
———————————————————————————————————————————————————————
B                                                      
  ASIAN         11 (13.9%)     9 (17.6%)    15 (19.5%) 
    range       23.0 - 40.0   27.0 - 48.0   26.0 - 58.0
  WHITE          5 (6.3%)      4 (7.8%)      8 (10.4%) 
    mean (sd)   35.0 (3.4)    39.0 (11.2)   32.2 (5.3) 
    range       31.0 - 39.0   24.0 - 48.0   26.0 - 42.0
———————————————————————————————————————————————————————

main footer

provenance footer

[[4]]
main title
subtitle

———————————————————————————————————————————————————————
                B: Placebo         C: Combination      
                     M             F             M     
———————————————————————————————————————————————————————
B                                                      
  ASIAN          7 (12.7%)    11 (16.7%)    14 (23.3%) 
    range       26.0 - 58.0   32.0 - 64.0   25.0 - 48.0
  WHITE          4 (7.3%)      5 (7.6%)      2 (3.3%)  
    mean (sd)   33.0 (9.8)    33.4 (6.5)    29.0 (4.2) 
    range       21.0 - 42.0   28.0 - 44.0   26.0 - 32.0
———————————————————————————————————————————————————————

main footer

provenance footer

rtables also supports page-by splits in its layouting framework, which declares that – regardless of rendering dimensions – pagination should occur between distinct levels of a variable. Each of these “pagination sections” have an additional title specific to the level, and are independently paginated for dimension as needed.

lyt2 <- basic_table(title = "main title", subtitles = "subtitle", main_footer = "main footer", prov_footer = "provenance footer") |>
  split_cols_by("ARM") |>
  split_rows_by("STRATA1", split_fun = keep_split_levels(c("A", "B")), page_by = TRUE, page_prefix = "Stratum") |>
  split_rows_by("RACE", split_fun = keep_split_levels(c("ASIAN", "WHITE"))) |>
  summarize_row_groups() |>
  analyze("AGE", afun = function(x, ...) in_rows("mean (sd)" = rcell(c(mean(x), sd(x)), format = "xx.x (xx.x)"),
                                                 "range" = rcell(range(x), format = "xx.x - xx.x")))

tbl2 <- build_table(lyt2, ex_adsl)

paginate_table(tbl2, lpp = 16)

$A1
main title
subtitle
Stratum: A

————————————————————————————————————————————————————————
               A: Drug X    B: Placebo    C: Combination
————————————————————————————————————————————————————————
ASIAN         22 (16.4%)    24 (17.9%)      19 (14.4%)  
  mean (sd)   31.9 (5.7)    35.2 (9.1)      35.2 (4.8)  
  range       24.0 - 43.0   23.0 - 62.0    28.0 - 47.0  
————————————————————————————————————————————————————————

main footer

provenance footer

$A2
main title
subtitle
Stratum: A

————————————————————————————————————————————————————————
               A: Drug X    B: Placebo    C: Combination
————————————————————————————————————————————————————————
WHITE          8 (6.0%)      7 (5.2%)        8 (6.1%)   
  mean (sd)   34.8 (5.1)    34.9 (7.5)      31.2 (6.8)  
  range       29.0 - 45.0   27.0 - 50.0    23.0 - 45.0  
————————————————————————————————————————————————————————

main footer

provenance footer

$B1
main title
subtitle
Stratum: B

————————————————————————————————————————————————————————
               A: Drug X    B: Placebo    C: Combination
————————————————————————————————————————————————————————
ASIAN         20 (14.9%)    23 (17.2%)      25 (18.9%)  
  mean (sd)   32.1 (6.9)    38.2 (9.8)      38.4 (8.8)  
  range       23.0 - 48.0   26.0 - 58.0    25.0 - 64.0  
————————————————————————————————————————————————————————

main footer

provenance footer

$B2
main title
subtitle
Stratum: B

————————————————————————————————————————————————————————
               A: Drug X    B: Placebo    C: Combination
————————————————————————————————————————————————————————
WHITE          10 (7.5%)     12 (9.0%)       7 (5.3%)   
  mean (sd)   35.9 (7.6)    32.5 (6.7)      32.1 (6.0)  
  range       24.0 - 48.0   21.0 - 42.0    26.0 - 44.0  
————————————————————————————————————————————————————————

main footer

provenance footer

3.3.2 flextable

The pagination of flextable objects let user control their position in relation to page breaks. It can prevent breaks between tables rows of a same group.

Function paginate() let you define this pagination, this feature is only available for RTF and Word outputs; see the documentation about pagination.

3.3.3 tables

Tables that span multiple pages are supported in PDF output using the LaTeX longtable package, specified when converting to LaTeX.

4.2.1 rtables

Adverse Events by ID

resetSession()
library(rtables)

s_events_patients <- function(x, labelstr, .N_col) {
  in_rows(
    "Patients with at least one event" =
      rcell(length(unique(x)) * c(1, 1 / .N_col), format = "xx (xx.xx%)"),

    "Total number of events" = rcell(length(x), format = "xx")
  )
}

table_count_per_id <- function(df, .N_col, termvar = "AEDECOD", idvar = "USUBJID") {

  x <- df[[termvar]]
  id <- df[[idvar]]

  counts <- table(x[!duplicated(paste0(id, x))])

  in_rows(
    .list = lapply(counts,
                   function(xi) rcell(c(xi, xi/.N_col), "xx (xx.xx%)")),
    .labels = names(counts)
  )
}

lyt <- basic_table(show_colcounts = TRUE) %>%
    split_cols_by("ARM") %>%
    analyze("USUBJID", afun = s_events_patients) %>%
    split_rows_by("AEBODSYS", child_labels = "visible", 
                  split_fun = trim_levels_in_group("AEDECOD"),
                  section_div = " ") %>%
    summarize_row_groups("USUBJID", cfun = s_events_patients) %>%
    analyze("AEDECOD", table_count_per_id, show_labels = "hidden", indent_mod = -1)

build_table(lyt, ex_adae, alt_counts_df = ex_adsl)

                                      A: Drug X      B: Placebo    C: Combination
                                       (N=134)        (N=134)         (N=132)    
—————————————————————————————————————————————————————————————————————————————————
Patients with at least one event     122 (91.04%)   123 (91.79%)    120 (90.91%) 
Total number of events                   609            622             703      
cl A.1                                                                           
  Patients with at least one event   78 (58.21%)    75 (55.97%)     89 (67.42%)  
  Total number of events                 132            130             160      
  dcd A.1.1.1.1                      50 (37.31%)    45 (33.58%)     63 (47.73%)  
  dcd A.1.1.1.2                      48 (35.82%)    48 (35.82%)     50 (37.88%)  
                                                                                 
cl B.1                                                                           
  Patients with at least one event   47 (35.07%)    49 (36.57%)     43 (32.58%)  
  Total number of events                  56             60              62      
  dcd B.1.1.1.1                      47 (35.07%)    49 (36.57%)     43 (32.58%)  
                                                                                 
cl B.2                                                                           
  Patients with at least one event   79 (58.96%)    74 (55.22%)     85 (64.39%)  
  Total number of events                 129            138             143      
  dcd B.2.1.2.1                      49 (36.57%)    44 (32.84%)     52 (39.39%)  
  dcd B.2.2.3.1                      48 (35.82%)    54 (40.30%)     51 (38.64%)  
                                                                                 
cl C.1                                                                           
  Patients with at least one event   43 (32.09%)    46 (34.33%)     43 (32.58%)  
  Total number of events                  55             63              64      
  dcd C.1.1.1.3                      43 (32.09%)    46 (34.33%)     43 (32.58%)  
                                                                                 
cl C.2                                                                           
  Patients with at least one event   35 (26.12%)    48 (35.82%)     55 (41.67%)  
  Total number of events                  48             53              65      
  dcd C.2.1.2.1                      35 (26.12%)    48 (35.82%)     55 (41.67%)  
                                                                                 
cl D.1                                                                           
  Patients with at least one event   79 (58.96%)    67 (50.00%)     80 (60.61%)  
  Total number of events                 127            106             135      
  dcd D.1.1.1.1                      50 (37.31%)    42 (31.34%)     51 (38.64%)  
  dcd D.1.1.4.2                      48 (35.82%)    42 (31.34%)     50 (37.88%)  
                                                                                 
cl D.2                                                                           
  Patients with at least one event   47 (35.07%)    58 (43.28%)     57 (43.18%)  
  Total number of events                  62             72              74      
  dcd D.2.1.5.3                      47 (35.07%)    58 (43.28%)     57 (43.18%)  

4.2.2 tern (+ rtables)

resetSession()
library(tern)

lyt <- basic_table(show_colcounts = TRUE) %>%
  split_cols_by(var = "ARM") %>%
  analyze_num_patients(
    vars = "USUBJID",
    .stats = c("unique", "nonunique"),
    .labels = c(
      unique = "Patients with at least one event",
      nonunique = "Total number of events"
    )
  ) %>%
  split_rows_by(
      "AEBODSYS",
      child_labels = "visible",
      split_fun = drop_split_levels,
      section_div = " "
  ) %>%
  summarize_num_patients(
    var = "USUBJID",
    .stats = c("unique", "nonunique"),
    .labels = c(
      unique = "Patients with at least one event",
      nonunique = "Total number of events"
    )
  ) %>%
  count_occurrences(vars = "AEDECOD", .indent_mods = -1L)

build_table(lyt, df = ex_adae, alt_counts_df = ex_adsl)

                                      A: Drug X    B: Placebo    C: Combination
                                       (N=134)       (N=134)        (N=132)    
———————————————————————————————————————————————————————————————————————————————
Patients with at least one event     122 (91.0%)   123 (91.8%)    120 (90.9%)  
Total number of events                   609           622            703      
cl A.1                                                                         
  Patients with at least one event   78 (58.2%)    75 (56.0%)      89 (67.4%)  
  Total number of events                 132           130            160      
  dcd A.1.1.1.1                      50 (37.3%)    45 (33.6%)      63 (47.7%)  
  dcd A.1.1.1.2                      48 (35.8%)    48 (35.8%)      50 (37.9%)  
                                                                               
cl B.1                                                                         
  Patients with at least one event   47 (35.1%)    49 (36.6%)      43 (32.6%)  
  Total number of events                 56            60              62      
  dcd B.1.1.1.1                      47 (35.1%)    49 (36.6%)      43 (32.6%)  
                                                                               
cl B.2                                                                         
  Patients with at least one event   79 (59.0%)    74 (55.2%)      85 (64.4%)  
  Total number of events                 129           138            143      
  dcd B.2.1.2.1                      49 (36.6%)    44 (32.8%)      52 (39.4%)  
  dcd B.2.2.3.1                      48 (35.8%)    54 (40.3%)      51 (38.6%)  
                                                                               
cl C.1                                                                         
  Patients with at least one event   43 (32.1%)    46 (34.3%)      43 (32.6%)  
  Total number of events                 55            63              64      
  dcd C.1.1.1.3                      43 (32.1%)    46 (34.3%)      43 (32.6%)  
                                                                               
cl C.2                                                                         
  Patients with at least one event   35 (26.1%)    48 (35.8%)      55 (41.7%)  
  Total number of events                 48            53              65      
  dcd C.2.1.2.1                      35 (26.1%)    48 (35.8%)      55 (41.7%)  
                                                                               
cl D.1                                                                         
  Patients with at least one event   79 (59.0%)    67 (50.0%)      80 (60.6%)  
  Total number of events                 127           106            135      
  dcd D.1.1.1.1                      50 (37.3%)    42 (31.3%)      51 (38.6%)  
  dcd D.1.1.4.2                      48 (35.8%)    42 (31.3%)      50 (37.9%)  
                                                                               
cl D.2                                                                         
  Patients with at least one event   47 (35.1%)    58 (43.3%)      57 (43.2%)  
  Total number of events                 62            72              74      
  dcd D.2.1.5.3                      47 (35.1%)    58 (43.3%)      57 (43.2%)  

4.2.3 gt

resetSession()

library(tidyverse)
library(gt)

ex_adsl <- formatters::ex_adsl
ex_adae <- formatters::ex_adae

header_n <- ex_adsl |> 
  dplyr::group_by(ARM) |> 
  dplyr::summarize(
    N = dplyr::n_distinct(USUBJID)
  ) 

col_lbls <- header_n |> 
  dplyr::transmute(
    ARMN = sprintf("%s  \n  (N=%i)", ARM, N)
  ) |> 
  dplyr::group_split(ARMN) 

sum_ex <- merge(ex_adae, header_n, by = "ARM") |> 
  dplyr::group_by(ARM) |> 
  dplyr::summarize(
    n_oe = dplyr::n_distinct(USUBJID),
    pct_oe = n_oe/mean(N),
    n_tot = dplyr::n(),
    .groups = "drop"
  )

sum_aebodsys <- merge(ex_adae, header_n, by = "ARM") |> 
  dplyr::group_by(ARM, AEBODSYS) |> 
  dplyr::summarize(
    n_oe = dplyr::n_distinct(USUBJID),
    pct_oe = n_oe/mean(N),
    n_tot = dplyr::n(),
    .groups = "drop"
  )

sum_aedecod <- merge(ex_adae, header_n, by = "ARM") |> 
  dplyr::group_by(ARM, AEBODSYS, AEDECOD) |> 
  dplyr::summarize(
    n_oe = dplyr::n_distinct(USUBJID),
    pct_oe = n_oe/mean(N),
    .groups = "drop"
  )

ex_tbl <- dplyr::bind_rows(sum_ex, sum_aebodsys, sum_aedecod) |> 
  tidyr::pivot_longer(cols = c(n_oe, n_tot), names_to = "lbl", values_to = "n") |> 
  dplyr::mutate(
    pct_oe = ifelse(lbl == "n_tot", NA_real_, pct_oe)
  ) |> 
  pivot_wider(id_cols = c(AEBODSYS, AEDECOD, lbl), names_from = ARM, values_from = c(n, pct_oe)) |> 
  dplyr::mutate(
    AEDECOD = forcats::fct_relevel(
      .f = dplyr::case_when(
        is.na(AEDECOD) & lbl == "n_tot" ~ "Total number of events",
        is.na(AEDECOD) & lbl == "n_oe" ~ "Patients with at least one event",
        TRUE ~ AEDECOD
      ),
      c("Patients with at least one event", "Total number of events"),
      after = 0
      ),
    AEBODSYS = forcats::fct_relevel(
      forcats::fct_na_value_to_level(
        AEBODSYS,
        level = " "
        ),
      " ",
      after = 0
    )
  ) |> 
  dplyr::filter(!(lbl == "n_tot" & !(AEDECOD %in% c("Patients with at least one event", "Total number of events")))) |> 
  dplyr::arrange(AEBODSYS, AEDECOD)

ex_tbl |> 
  gt(
    rowname_col = "AEDECOD",
    groupname_col = "AEBODSYS"
  ) |> 
  cols_hide(columns = "lbl") |> 
  fmt_percent(
    columns = starts_with("pct"),
    decimals = 1
  ) |> 
  cols_merge_n_pct(
    col_n = "n_A: Drug X",
    col_pct = "pct_oe_A: Drug X"
  ) |> 
  cols_merge_n_pct(
    col_n = "n_B: Placebo",
    col_pct = "pct_oe_B: Placebo"
  ) |> 
  cols_merge_n_pct(
    col_n = "n_C: Combination",
    col_pct = "pct_oe_C: Combination"
  ) |> 
  cols_label(
    "n_A: Drug X" = md(col_lbls[[1]]),
    "n_B: Placebo" = md(col_lbls[[2]]),
    "n_C: Combination" = md(col_lbls[[3]])
  ) |> 
  cols_align(
    columns = 3:9,
    align = "center"
  ) |> 
  cols_align(
    columns = 1:2,
    align = "left"
  ) |> 
  cols_width(
    .list = list(
      1:2 ~ px(250),
      3:9 ~ px(120)
    )
  ) |> 
  tab_stub_indent(
    rows = 2:18,
    indent = 3
  )

4.2.4 tables

The tables package normally generates tables from single datasets, while this kind of table requires information from two: adsl and ex_adae. One way to handle this would be to add the adsl patient count information to a copy of the ex_adae table. In this code we use a different approach: we generate one table of patient counts to produce the heading lines, and a second table with the adverse event data, then use rbind() to combine the two tables.

resetSession()

library(tables)
table_options(doCSS = TRUE)

ex_adae <- formatters::ex_adae

subject_counts <- table(adsl$ARM)

countpercentid <- function(num, ARM) {
  n <- length(unique(num))
  if (n == 0) pct <- 0
  else        pct <- 100*n/subject_counts[ARM[1]]
  sprintf("%d (%.2f%%)", 
          length(unique(num)), 
          pct)
}

count <- function(x) sprintf("(N=%d)", length(x))

heading <- tabular(Heading("")*1*
                     Heading("")*count ~ 
                   Heading()*ARM, data = adsl)

body <- tabular( Heading("Patients with at least one event")*1*
                   Heading("")*countpercentid*Arguments(ARM = ARM)*
                   Heading()*USUBJID +
                 Heading("Total number of events")*1*Heading("")*1 +
                 Heading()*AEBODSYS*
                   (Heading("Patients with at least one event")*
                      Percent(denom = ARM, fn = countpercentid)*
                      Heading()*USUBJID +
                    Heading("Total number of events")*1 +
                    Heading()*AEDECOD*DropEmpty(which = "row")*
                      Heading()*Percent(denom = ARM, fn = countpercentid)*
                      Heading()*USUBJID) ~ 
                 Heading()*ARM, 
                 data = ex_adae )

tab <- rbind(heading, body)
useGroupLabels(tab, indent = "&emsp;", extraLines = 1)

4.2.5 flextable

By using tables::tabular() to create a table and then converting it to a flextable using as_flextable(), you can take advantage of the convenience and flexibility provided by the tables package while still benefiting from the formatting capabilities of flextable.

library(flextable)
as_flextable(body, spread_first_col = TRUE, add_tab = TRUE) |>
  align(j = 1, part = "all", align = "left") |> 
  padding(padding = 4, part = "all") |> 
  add_header_row(
    values = c("", fmt_header_n(subject_counts, newline = FALSE)),
    top = FALSE) |> 
  hline(i = 1, part = "header", border = fp_border_default(width = 0))

4.2.6 tidytlg

resetSession()
library(dplyr)
library(tidytlg)

adsl <- formatters::ex_adsl
adae <- formatters::ex_adae %>% 
  mutate(TRTEMFL = "Y")

# Create analysis population counts
tbl1 <- freq(adsl,
             rowvar = "SAFFL",
             colvar = "ARM",
             statlist = statlist("n"),
             rowtext = "Analysis Set: Safety Population",
             subset = SAFFL == "Y")

# Create counts (percentages) for patients with at least one event
tbl2 <- freq(adae,
             denom_df = adsl,
             rowvar = "TRTEMFL",
             colvar = "ARM",
             statlist = statlist("n (x.x%)"),
             rowtext = "Patients with at least one event",
             subset = TRTEMFL == "Y")

# Create counts (percentages) of AE by AEBODSYS and AEDECOD
tbl3a <- nested_freq(adae,
                    denom_df = adsl,
                    rowvar = "AEBODSYS*AEDECOD",
                    colvar = "ARM",
                    statlist = statlist("n (x.x%)"))

# Create total event counts by AEBODSYS
tbl3b <- freq(adae,
              rowvar = "AEBODSYS",
              colvar = "ARM",
              statlist = statlist("n", distinct = FALSE)) %>% 
  rename(AEBODSYS = label) %>% 
  mutate(label = "Total number of events",
         nested_level = 0)

# interleave tbl3a and tbl3b by AEBODSYS
tbl3 <- bind_rows(tbl3a, tbl3b) %>% 
  arrange(AEBODSYS, nested_level)

# combine analysis results together
tbl <- bind_table(tbl1, tbl2, tbl3) %>% 
  select(-AEBODSYS)

# output the analysis results
gentlg(huxme       = tbl,
       format      = "HTML",
       print.hux = FALSE,
       file        = "Table x.x.x.x",
       orientation = "portrait",
       title = "Adverse Events Summary - Safety Analysis Set",
       colheader = c("","A: Drug X","B: Placebo","C: Combination"))

4.2.7 tfrmt

Rather than starting with an ADaM, tfrmt assumes users will start with an ARD (Analysis Results Dataset), because of this, making this table will be split into two parts, first to make the ARD and second to format the table.

resetSession()
library(tidyverse)
library(tfrmt)

# Make ARD 
ex_adsl <- formatters::ex_adsl
ex_adae <- formatters::ex_adae 

big_n <- ex_adsl |> 
  dplyr::group_by(ARM) |> 
  dplyr::summarize(
    N = dplyr::n_distinct(USUBJID)
  ) 

adae_with_n <- ex_adae |> 
  dplyr::left_join(big_n, by = "ARM")

calc_tot_and_any <- function(.data){
  .data |>
    dplyr::reframe(
      n_subj = n_distinct(USUBJID),
      pct_subj = n_subj/N,
      n_evnts = n()
    ) |> 
    dplyr::distinct() |>
    tidyr::pivot_longer(c("n_subj", "pct_subj", "n_evnts")) |> 
    dplyr::mutate(label = dplyr::case_when(
      name %in% c("n_subj", "pct_subj") ~ "Patients with at least one event",
      name == "n_evnts" ~ "Total number of events"
    ))
}

overall <- adae_with_n |> 
  dplyr::group_by(ARM) |> 
  calc_tot_and_any() |>
  dplyr:: mutate(AEBODSYS = label)

bdysys_overall <- adae_with_n |> 
  dplyr::group_by(ARM, AEBODSYS) |> 
  calc_tot_and_any()

aeterm_sum <- adae_with_n |> 
  dplyr::group_by(ARM, AEBODSYS, AETERM) |> 
  dplyr::reframe(
      n_subj = n_distinct(USUBJID),
      pct_subj = n_subj/N) |> 
  dplyr::distinct() |>
  tidyr::pivot_longer(ends_with("subj")) |> 
  dplyr::rename(label = AETERM)

header_n <- big_n |> 
  dplyr::rename(value = N) |> 
  dplyr::mutate(name = "header_n")

ae_ard <- dplyr::bind_rows(
  overall, 
  bdysys_overall,
  aeterm_sum,
  header_n
)

## Format Table 
tfrmt(
  column = ARM,
  group = c("AEBODSYS"),
  param = name,
  value = value,
  label = label, 
) |>
  # Then we cam combine it with an n percent template 
tfrmt_n_pct(n = "n_subj",
            pct = "pct_subj",
  pct_frmt_when = frmt_when("==1" ~ "", 
                            ">.99" ~ "(>99%)", 
                            "==0" ~ "", 
                            "<.01" ~ "(<1%)", 
                            "TRUE" ~ frmt("(xx.x%)", transform = ~.*100))
  ) |>
  #Finally we are going to add some additional formatting
  tfrmt(
    body_plan = body_plan(
      frmt_structure("n_evnts" = frmt("XXX"))
    ),
    big_n = big_n_structure("header_n"),
    # Aligning on decimal places and spaces
    col_style_plan = col_style_plan(
      col_style_structure(col = matches("[A-Z]:.*"),
                          align = c(".", " "))
    )
  ) |> 
  print_to_gt(ae_ard)

4.3.1 Data and models used throughout

resetSession()
library(dplyr)
library(tidyr)
library(stringr)
library(purrr)
library(survival)

data("cadaette", package = "random.cdisc.data")
head(cadaette)

# A tibble: 6 × 66
  STUDYID USUBJID     SUBJID SITEID   AGE AGEU  SEX   RACE  ETHNIC COUNTRY DTHFL
  <chr>   <chr>       <chr>  <chr>  <int> <fct> <fct> <fct> <fct>  <fct>   <fct>
1 AB12345 AB12345-BR… id-105 BRA-1     38 YEARS M     BLAC… HISPA… BRA     N    
2 AB12345 AB12345-BR… id-105 BRA-1     38 YEARS M     BLAC… HISPA… BRA     N    
3 AB12345 AB12345-BR… id-105 BRA-1     38 YEARS M     BLAC… HISPA… BRA     N    
4 AB12345 AB12345-BR… id-105 BRA-1     38 YEARS M     BLAC… HISPA… BRA     N    
5 AB12345 AB12345-BR… id-105 BRA-1     38 YEARS M     BLAC… HISPA… BRA     N    
6 AB12345 AB12345-BR… id-105 BRA-1     38 YEARS M     BLAC… HISPA… BRA     N    
# ℹ 55 more variables: INVID <chr>, INVNAM <chr>, ARM <fct>, ARMCD <fct>,
#   ACTARM <fct>, ACTARMCD <fct>, TRT01P <fct>, TRT01A <fct>, TRT02P <fct>,
#   TRT02A <fct>, REGION1 <fct>, STRATA1 <fct>, STRATA2 <fct>, BMRKR1 <dbl>,
#   BMRKR2 <fct>, ITTFL <fct>, SAFFL <fct>, BMEASIFL <fct>, BEP01FL <fct>,
#   AEWITHFL <fct>, RANDDT <date>, TRTSDTM <dttm>, TRTEDTM <dttm>,
#   TRT01SDTM <dttm>, TRT01EDTM <dttm>, TRT02SDTM <dttm>, TRT02EDTM <dttm>,
#   AP01SDTM <dttm>, AP01EDTM <dttm>, AP02SDTM <dttm>, AP02EDTM <dttm>, …

adtte <- cadaette %>% 
    dplyr::filter(PARAMCD == "AETTE2", SAFFL == "Y")

Cox Proportional Hazard fit:

cph <- coxph(Surv(AVAL, CNSR==0) ~ TRT01A + STRATA1, ties = "exact", data = adtte)

Kaplan-Meier Model

surv_tbl <- as.data.frame(summary(survfit(Surv(AVAL, CNSR==0) ~ TRT01A,
                                          data = adtte, conf.type = "log-log"))$table) %>%
    dplyr::mutate(TRT01A = factor(str_remove(row.names(.), "TRT01A="),
                                  levels = levels(adtte$TRT01A)),
                  ind = FALSE)
mn_footer_txt <- c("Serious adverse events are defined as (...). All p-values are exploratory.",
                   "Hazard ratios are from a stratified Cox model of serious adverse event hazard rate,",
                   "with terms for treatment groups and strata1. Ties were handled using the exact",
                   "method. Hazard ratios of Placebo Combination over Drug X are presented, an",
                   "HR < 1 denotes improvement compared to Drug X.")

stitle_txt <- c("x.x.x: Time to First Serious Adverse Event",
                "Table x.x.x.x: Safety Endpoint - Safety Analysis Set")
.kmState <- currentState()

4.3.2 rtables

resetSession(.kmState)
library(rtables)

## this will be properly exported in the next release of rtables
RefFootnote <- rtables:::RefFootnote

cnsr_counter <- function(df, .var, .N_col) {
    x <- df[!duplicated(df$USUBJID), .var]
    x <- x[x != "__none__"]
    lapply(table(x), function(xi) rcell(xi*c(1, 1/.N_col), format = "xx (xx.xx%)"))
}
            
a_count_subjs <- function(x, .N_col) {
    in_rows("Subjects with Adverse Events n (%)" = rcell(length(unique(x)) * c(1, 1 / .N_col),
                                                                           format = "xx (xx.xx%)"))
}

a_cph <- function(df, .var, .in_ref_col, .ref_full, full_cox_fit) {
    if(.in_ref_col) {
        ret <- replicate(3, list(rcell(NULL)))
    } else {
        curtrt <- df[[.var]][1]
        coefs <- coef(full_cox_fit)
        sel_pos <- grep(curtrt, names(coefs), fixed = TRUE)
        hrval <- exp(coefs[sel_pos])
        hrvalret <- rcell(hrval, format = "xx.x")
        sdf <- survdiff(Surv(AVAL, CNSR==0) ~ TRT01A + STRATA1,
                        data = rbind(df, .ref_full))
        pval <- (1-pchisq(sdf$chisq, length(sdf$n)-1))/2
        ci_val <- exp(unlist(confint(full_cox_fit)[sel_pos,]))
        ret <- list(rcell(hrval, format = "xx.x"),
                    rcell(ci_val, format = "(xx.x, xx.x)"),
                    rcell(pval, format = "x.xxxx | (<0.0001)"))
    }
    in_rows(.list = ret, .names = c("Hazard ratio",
                                    "95% confidence interval",
                                    "p-value (one-sided stratified log rank)"))
}

a_tte <- function(df, .var,  kp_table) {
    ind <- grep(df[[.var]][1], row.names(kp_table), fixed = TRUE)
    minmax <- range(df[["AVAL"]])

    mm_val_str <- format_value(minmax, format = "xx.x, xx.x")
    rowfn <- list()
    in_rows(Median = kp_table[ind, "median", drop = TRUE],
            "95% confidence interval" = unlist(kp_table[ind, c("0.95LCL", "0.95UCL")]),
            "Min Max" = mm_val_str,
            .formats = c("xx.xx",
                         "xx.xx - xx.xx",
                         "xx"), .cell_footnotes = list(NULL, NULL, list(RefFootnote("Denotes censoring", index = 0L, symbol = "*"))))
}
            

adtte2 <- adtte |>
    mutate(CNSDTDSC = ifelse(CNSDTDSC == "", "__none__", CNSDTDSC))

lyt <- basic_table(show_colcounts = TRUE,
                   title = "x.x: Safety Data",
                   subtitles = stitle_txt,
                   main_footer = mn_footer_txt,
                   prov_footer = "Source: ADTTE DDMMYYYY hh:mm; Listing x.xx; SDTM package: DDMMYYYY") |>
    split_cols_by("ARM", ref_group = "A: Drug X") |>
    analyze("USUBJID", a_count_subjs, show_labels = "hidden") |>
    analyze("CNSDTDSC", cnsr_counter, var_labels = "Censored Subjects", show_labels = "visible") |>
    analyze("ARM", a_cph, extra_args = list(full_cox_fit = cph), show_labels = "hidden") |>
    analyze("ARM", a_tte, var_labels = "Time to first adverse event", show_labels = "visible",
            extra_args = list(kp_table = surv_tbl),
            table_names = "kapmeier")

tbl_tte <- build_table(lyt, adtte2)

fnotes_at_path(tbl_tte, c("ma_USUBJID_CNSDTDSC_ARM_kapmeier", "kapmeier")) <- "Product-limit (Kaplan-Meier) estimates."
tbl_tte

x.x: Safety Data
x.x.x: Time to First Serious Adverse Event
Table x.x.x.x: Safety Endpoint - Safety Analysis Set

————————————————————————————————————————————————————————————————————————————————————————
                                            A: Drug X      B: Placebo     C: Combination
                                             (N=134)         (N=134)         (N=132)    
————————————————————————————————————————————————————————————————————————————————————————
Subjects with Adverse Events n (%)        134 (100.00%)   134 (100.00%)   132 (100.00%) 
Censored Subjects                                                                       
  Clinical Cut Off                         10 (7.46%)       4 (2.99%)      14 (10.61%)  
  Completion or Discontinuation            13 (9.70%)       3 (2.24%)      16 (12.12%)  
  End of AE Reporting Period               22 (16.42%)      4 (2.99%)      14 (10.61%)  
Hazard ratio                                                   1.5             1.1      
95% confidence interval                                    (1.1, 1.9)       (0.8, 1.5)  
p-value (one-sided stratified log rank)                      0.0208           0.4619    
Time to first adverse event {1}                                                         
  Median                                      0.39            0.37             0.26     
  95% confidence interval                  0.23 - 0.60     0.25 - 0.46     0.18 - 0.34  
  Min Max                                 0.0, 3.0 {*}    0.0, 3.0 {*}     0.0, 3.0 {*} 
————————————————————————————————————————————————————————————————————————————————————————

{1} - Product-limit (Kaplan-Meier) estimates.
{*} - Denotes censoring
————————————————————————————————————————————————————————————————————————————————————————

Serious adverse events are defined as (...). All p-values are exploratory.
Hazard ratios are from a stratified Cox model of serious adverse event hazard rate,
with terms for treatment groups and strata1. Ties were handled using the exact
method. Hazard ratios of Placebo Combination over Drug X are presented, an
HR < 1 denotes improvement compared to Drug X.

Source: ADTTE DDMMYYYY hh:mm; Listing x.xx; SDTM package: DDMMYYYY

4.3.3 tern (+rtables)

tern encapsulates the specific statistical choices used by Roche. In particulate, its implementation of the Cox pairwise analysis does not implement the one-tailed p-value strategy used in the rest of this chapter.

We will first showcaes the pure tern solution, which has different p-values for this reason, and then implement a hybrid tern + explicit rtables solution which fully recreates the exact table generated by other systems.

resetSession(.kmState)
library(tern)

## this will be properly exported in the next release of rtables
RefFootnote <- rtables:::RefFootnote

adtte3 <- adtte
adtte3$is_event <- adtte$CNSR == 0
adtte3$CNSDTDSC[adtte$CNSDTDSC == ""] <- NA



lyt1 <- basic_table(show_colcounts = TRUE,
                   title = "x.x: Safety Data",
                   subtitles = stitle_txt,
                   main_footer = mn_footer_txt,
                   prov_footer = "Source: ADTTE DDMMYYYY hh:mm; Listing x.xx; SDTM package: DDMMYYYY") |>
  split_cols_by("TRT01A", ref_group = "A: Drug X") |>
  count_values(
    "STUDYID",
    values = "AB12345",
    .stats = "count_fraction",
    .labels = c(count_fraction = "Subjects with Adverse Events n (%)")
  ) |>
  count_occurrences(
    "CNSDTDSC",
    var_labels = "Censored Subjects",
    show_labels = "visible",
    .formats = c(count_fraction = "xx.xx (xx.xx%)")
  ) |>
  coxph_pairwise(
    vars = "AVAL",
    is_event = "is_event",
    control = control_coxph(pval_method = "log-rank", ties = "exact"),
    strat = "STRATA1",
    .stats = c("hr", "hr_ci", "pvalue"),
    .formats = c(hr = "xx.x", hr_ci = "(xx.x, xx.x)", pvalue = "xx.xxxx"),
    .labels = c(hr = "Hazard ratio", hr_ci = "95% confidence interval", pvalue = "p-value (stratified log rank)"),
    show_labels = "hidden",
    table_names = "coxph"
  ) |>
  surv_time(
    vars = "AVAL",
    is_event = "is_event",
    control = control_surv_time(conf_type = "log-log"),
    .stats = c("median", "median_ci", "range"),
    .formats = c(median = "xx.xx", median_ci = "xx.xx - xx.xx", range = "xx.x, xx.x"),
    .labels = c(median_ci = "95% confidence interval", range = "Min Max"),
    .indent_mods = c(median_ci = 0L),
    var_labels = "Time to first adverse event"
  )

tbl_tte_tern <- build_table(lyt = lyt1, df = adtte3)

fnotes_at_path(tbl_tte_tern, c("ma_STUDYID_CNSDTDSC_coxph_AVAL", "AVAL")) <- "Product-limit (Kaplan-Meier) estimates."

fnote <- RefFootnote("Censored.", index = 0L, symbol = "^")

for(pth in col_paths(tbl_tte_tern)) {
    fnotes_at_path(tbl_tte_tern,
                   rowpath = c("ma_STUDYID_CNSDTDSC_coxph_AVAL", "AVAL", "range"),
                   colpath = pth) <- fnote
}
tbl_tte_tern

x.x: Safety Data
x.x.x: Time to First Serious Adverse Event
Table x.x.x.x: Safety Endpoint - Safety Analysis Set

————————————————————————————————————————————————————————————————————————————————————
                                       A: Drug X       B: Placebo     C: Combination
                                        (N=134)          (N=134)         (N=132)    
————————————————————————————————————————————————————————————————————————————————————
Subjects with Adverse Events n (%)   134 (100.00%)    134 (100.00%)   132 (100.00%) 
Censored Subjects                                                                   
  Clinical Cut Off                   10.00 (7.46%)    4.00 (2.99%)    14.00 (10.61%)
  Completion or Discontinuation      13.00 (9.70%)    3.00 (2.24%)    16.00 (12.12%)
  End of AE Reporting Period         22.00 (16.42%)   4.00 (2.99%)    14.00 (10.61%)
Hazard ratio                                               1.5             1.1      
95% confidence interval                                (1.2, 2.0)       (0.8, 1.5)  
p-value (stratified log rank)                            0.0023           0.6027    
Time to first adverse event {1}                                                     
  Median                                  0.39            0.37             0.26     
  95% confidence interval             0.23 - 0.60      0.25 - 0.46     0.18 - 0.34  
  Min Max                             0.0, 3.0 {^}    0.0, 3.0 {^}     0.0, 3.0 {^} 
————————————————————————————————————————————————————————————————————————————————————

{1} - Product-limit (Kaplan-Meier) estimates.
{^} - Censored.
————————————————————————————————————————————————————————————————————————————————————

Serious adverse events are defined as (...). All p-values are exploratory.
Hazard ratios are from a stratified Cox model of serious adverse event hazard rate,
with terms for treatment groups and strata1. Ties were handled using the exact
method. Hazard ratios of Placebo Combination over Drug X are presented, an
HR < 1 denotes improvement compared to Drug X.

Source: ADTTE DDMMYYYY hh:mm; Listing x.xx; SDTM package: DDMMYYYY

We now create the hybrid table where we utilize a custom analysis function to recreate the one-sided p-values while using tern for the rest of the table structure.

a_cph <- function(df, .var, .in_ref_col, .ref_full, full_cox_fit) {
  if(.in_ref_col) {
    ret <- replicate(3, list(rcell(NULL)))
  } else {
    curtrt <- df[[.var]][1]
    coefs <- coef(full_cox_fit)
    sel_pos <- grep(curtrt, names(coefs), fixed = TRUE)
    hrval <- exp(coefs[sel_pos])
    hrvalret <- rcell(hrval, format = "xx.x")
    sdf <- survival::survdiff(Surv(AVAL, CNSR==0) ~ TRT01A + STRATA1,
                              data = rbind(df, .ref_full))
    pval <- (1-pchisq(sdf$chisq, length(sdf$n)-1))/2
    ci_val <- exp(unlist(confint(full_cox_fit)[sel_pos,]))
    ret <- list(rcell(hrval, format = "xx.x"),
                rcell(ci_val, format = "(xx.x, xx.x)"),
                rcell(pval, format = "x.xxxx | (<0.0001)"))
  }
  in_rows(.list = ret, .names = c("Hazard ratio",
                                  "95% confidence interval",
                                  "p-value (one-sided stratified log rank)"))
}

lyt2 <- basic_table(show_colcounts = TRUE,
                   title = "x.x: Safety Data",
                   subtitles = stitle_txt,
                   main_footer = mn_footer_txt,
                   prov_footer = "Source: ADTTE DDMMYYYY hh:mm; Listing x.xx; SDTM package: DDMMYYYY") |>
  split_cols_by("TRT01A", ref_group = "A: Drug X") |>
  count_values(
    "STUDYID",
    values = "AB12345",
    .stats = "count_fraction",
    .labels = c(count_fraction = "Subjects with Adverse Events n (%)")
  ) |>
  count_occurrences(
    "CNSDTDSC",
    var_labels = "Censored Subjects",
    show_labels = "visible",
    .formats = c(count_fraction = "xx.xx (xx.xx%)")
  ) |>
  analyze("ARM", a_cph, extra_args = list(full_cox_fit = cph), show_labels = "hidden") |>
  surv_time(
    vars = "AVAL",
    is_event = "is_event",
    control = control_surv_time(conf_type = "log-log"),
    .stats = c("median", "median_ci", "range"),
    .formats = c(median = "xx.xx", median_ci = "xx.xx - xx.xx", range = "xx.x, xx.x"),
    .labels = c(median_ci = "95% confidence interval", range = "Min Max"),
    .indent_mods = c(median_ci = 0L),
    var_labels = "Time to first adverse event"
  )

tbl_tte_tern2 <- build_table(lyt = lyt2, df = adtte3)

fnotes_at_path(tbl_tte_tern2, c("ma_STUDYID_CNSDTDSC_ARM_AVAL", "AVAL")) <- "Product-limit (Kaplan-Meier) estimates."

fnote <- RefFootnote("Denotes censoring.", index = 0L, symbol = "*")

for(pth in col_paths(tbl_tte_tern2)) {
    fnotes_at_path(tbl_tte_tern2,
                   rowpath = c("ma_STUDYID_CNSDTDSC_ARM_AVAL", "AVAL", "range"),
                   colpath = pth) <- fnote
}
tbl_tte_tern2

x.x: Safety Data
x.x.x: Time to First Serious Adverse Event
Table x.x.x.x: Safety Endpoint - Safety Analysis Set

—————————————————————————————————————————————————————————————————————————————————————————
                                            A: Drug X       B: Placebo     C: Combination
                                             (N=134)          (N=134)         (N=132)    
—————————————————————————————————————————————————————————————————————————————————————————
Subjects with Adverse Events n (%)        134 (100.00%)    134 (100.00%)   132 (100.00%) 
Censored Subjects                                                                        
  Clinical Cut Off                        10.00 (7.46%)    4.00 (2.99%)    14.00 (10.61%)
  Completion or Discontinuation           13.00 (9.70%)    3.00 (2.24%)    16.00 (12.12%)
  End of AE Reporting Period              22.00 (16.42%)   4.00 (2.99%)    14.00 (10.61%)
Hazard ratio                                                    1.5             1.1      
95% confidence interval                                     (1.1, 1.9)       (0.8, 1.5)  
p-value (one-sided stratified log rank)                       0.0208           0.4619    
Time to first adverse event {1}                                                          
  Median                                       0.39            0.37             0.26     
  95% confidence interval                  0.23 - 0.60      0.25 - 0.46     0.18 - 0.34  
  Min Max                                  0.0, 3.0 {*}    0.0, 3.0 {*}     0.0, 3.0 {*} 
—————————————————————————————————————————————————————————————————————————————————————————

{1} - Product-limit (Kaplan-Meier) estimates.
{*} - Denotes censoring.
—————————————————————————————————————————————————————————————————————————————————————————

Serious adverse events are defined as (...). All p-values are exploratory.
Hazard ratios are from a stratified Cox model of serious adverse event hazard rate,
with terms for treatment groups and strata1. Ties were handled using the exact
method. Hazard ratios of Placebo Combination over Drug X are presented, an
HR < 1 denotes improvement compared to Drug X.

Source: ADTTE DDMMYYYY hh:mm; Listing x.xx; SDTM package: DDMMYYYY

4.3.4 Cell Value Derivation for gt of Time to Event Analysis

Our standard TTE table consists of (a derivation of) four main parts:

1. Descriptive stats including the number of subjects with an event, number of subjects censored and censoring reasons
2. Hazard ratio with corresponding 95% CI from a (stratified) Cox model and a p-value from a stratified log rank test
3. Median time to event Kaplan-Meier analysis
4. Number of patients at risk at specified visits from Kaplan-Meier analysis (omitted here).

resetSession(.kmState)
library(gt)
### Subject Count with events

## surv_tbl calculated above


subj_count <- surv_tbl  |> 
  dplyr::mutate(pct = sprintf("%i (%5.1f)", events, 100*events/records),
                label = "Number of subjects with serious adverse event, n (%)") |>
  dplyr::select(label, TRT01A, pct) |>
  tidyr::pivot_wider(id_cols = label, names_from = TRT01A, values_from = pct) |>
  dplyr::mutate(ind = FALSE)

# Number of censored subjects

cnsrd_subj_full <- surv_tbl |>
  dplyr::mutate(pct = sprintf("%i (%4.1f)", records-events, 100*(records-events)/records),
                CNSDTDSC = "Number of censored subjects, n (%)") |> 
  dplyr::select(CNSDTDSC, TRT01A, pct)
  
cnsrd_subj <- adtte |> 
  dplyr::group_by(TRT01A) |>
  dplyr::mutate(CNSR = CNSR/n()) |>
  dplyr::ungroup() |>
  dplyr::filter(CNSR != 0) |>
  dplyr::group_by(TRT01A, CNSDTDSC) |>
  dplyr::summarise(pct = sprintf("%i (%4.1f)", sum(CNSR != 0), 100*sum(CNSR)), .groups = "drop") |>
  dplyr::bind_rows(cnsrd_subj_full) |> 
  tidyr::pivot_wider(id_cols = CNSDTDSC, names_from = TRT01A, values_from = pct) |>
  dplyr::rename(label = CNSDTDSC) |>
  dplyr::mutate(ind = label != "Number of censored subjects, n (%)") |>
  dplyr::arrange(ind)