In GSK-Biostatistics/tlang: Applies Display Metadata to Analysis Results Datasets

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

library(tfrmt)
library(tibble)
library(dplyr)
library(tidyr)

Displaying subject totals ("Big N's") at the top of a table is a very common and necessary part of building tables for clinical trials. {tfrmt} allows you to do this via the big_n argument, which allows you to use a big_n_structure object to specify which values in the ARD should be added to the columns and how they should be formatted.

Overall Big Ns

To dynamically add big N's to column labels, the information needs to be included in the ARD. As a simple example, let's take a small demographics table that just has Age and Sex for Placebo, Treatment and Total. Below is our ARD prior to incorporating the big Ns:

data <- tibble(Group = rep(c("Age (y)", "Sex", "Age (y)", "Sex"), c(3, 3, 6,12)),
               Label = rep(c("n", "Mean (SD)", "Male","Female"), c(6, 6,6,6)),
               Column = rep(c("Placebo", "Treatment", "Total"), times = 8),
               Param = rep(c("n", "mean", "sd", "n", "pct", "n", "pct"),  c(6, 3, 3, 3,3,3,3)),
               Value = c(15,13,28,14,13,27,73.56, 74.231,71.84,9.347,7.234,8.293,8,7,15,8/14,7/13,15/27,6,6,12,6/14,6/13,12/27
               )
) %>%
  # Note because tfrmt only does rounding we will need to have the percents multiplied by 100
  mutate(Value = case_when(Param == "pct" ~ Value * 100,
                           TRUE ~ Value),
         ord1 = if_else(Group == "Age (y)", 1, 2),
         ord2 = if_else(Label == "n", 1, 2))
data

In this study we have 30 subjects per treatment arm for a total of 60 subjects. Because the ARD is 1 row per unique value, our Ns will contribute 3 rows to the ARD. To make the N's go to the right place, we need to supply information about the column column values. The column column values should map to those already available in the ARD. Then we can combine this information with the data above to make the full ARD for this table.

big_ns <- tibble(Column = c("Placebo", "Treatment", "Total"),
                 Param = "bigN",
                 Value = c(30,30,60))


data <- bind_rows(data, big_ns)

Now we can make our tfrmt as normal, but this time we are going to add a big_n_structure to big_n. Here we will specify that values with the parameter of "bigN" should be appended to the column labels and they should be formatted as "N = XX" and on a separate line.

tfrmt(
  group = Group,
  label = Label,
  column = Column,
  value = Value,
  param = Param,
  sorting_cols = c(ord1, ord2),
  body_plan = body_plan(
    frmt_structure(group_val = ".default",
                   label_val = ".default",
                   frmt_combine("{n} {pct}",
                                n = frmt("X"),
                                pct = frmt("(xx.x%)", missing = " ")
                   )
    ),
    frmt_structure(group_val = "Age (y)", label_val = "Mean (SD)",
                   frmt_combine("{mean} ({sd})",
                                mean = frmt("XX.X"),
                                sd = frmt("x.xx")
                   )
    ),
    frmt_structure(group_val = ".default", label_val = "n", frmt("xx"))
  ),
  col_plan = col_plan(everything(), -starts_with("ord"), "Total"),
  row_grp_plan = row_grp_plan(
    row_grp_structure(group_val = ".default", element_block(post_space = " "))
  ),
  big_n = big_n_structure(param_val = "bigN", n_frmt = frmt("\nN = xx"))
) %>% 
  print_to_gt(data)

No matter how complicated the table, this feature will work as long as the ARD is set up appropriately with the correct column column values on the big N rows. For tables with spanning columns (i.e. multiple column columns), the big N will be appended to the lowest level column column value that is supplied (based on the order of the column variables supplied to the tfrmt). Let's look at this table, which has 3 layers of column spanning:

  data <- tribble(
    ~group,     ~label,        ~span2,  ~span1,     ~my_col,    ~parm,   ~val,
    "g1", "rowlabel1",  "column cols", "cols 1,2", "col1"  ,  "value",    1,
    "g1", "rowlabel1",  "column cols", "cols 1,2", "col2"  ,  "value",    1,
    "g1", "rowlabel1",             NA,         NA, "mycol3",  "value",    1,
    "g1", "rowlabel1",  "column cols", "col 4"   , "col4"  ,  "value",    1,
    "g1", "rowlabel1",             NA,         NA, "mycol5",  "value",    1,
    "g1", "rowlabel2",  "column cols", "cols 1,2", "col1"  ,  "value",    2,
    "g1", "rowlabel2",  "column cols", "cols 1,2", "col2"  ,  "value",    2,
    "g1", "rowlabel2",             NA,        NA , "mycol3",  "value",    2,
    "g1", "rowlabel2",  "column cols", "col 4"   , "col4"  ,  "value",    2,
    "g1", "rowlabel2",             NA,         NA, "mycol5",  "value",    2,
    "g2", "rowlabel3",  "column cols", "cols 1,2", "col1"  ,  "value",    3,
    "g2", "rowlabel3",  "column cols", "cols 1,2", "col2"  ,  "value",    3,
    "g2", "rowlabel3",             NA,         NA, "mycol3",  "value",    3,
    "g2", "rowlabel3",  "column cols", "col 4"   , "col4"  ,  "value",    3,
    "g2", "rowlabel3",             NA,         NA, "mycol5",  "value",    3,
  )



spanning_tfrmt<- tfrmt(
    group = group,
    label = label,
    param = parm,
    value = val,
    column = c(span2, span1, my_col),
    body_plan = body_plan(
      frmt_structure(group_val = ".default", label_val = ".default", frmt("x"))
    ),
    col_plan = col_plan(
      group,
      label,
      starts_with("col")
    )
  )

print_to_gt(spanning_tfrmt, data)

If we just want to put the big N on the highest "column cols", we need to add a row to the ARD where span2 equals "column cols" and all other column columns are missing.

with_big_n_data <- tribble(
  ~group,     ~label,        ~span2,  ~span1,     ~my_col,    ~parm,   ~val,
      NA,        NA,  "column cols",     NA,       NA  ,  "bigN",    18,
) %>% 
  bind_rows(data)

#Now we can add the big_n to the tfrmt from before 
n_span_tfrmt <- spanning_tfrmt %>% 
  tfrmt(big_n = big_n_structure(param_val = "bigN", n_frmt = frmt("\nN = xx"))) 

print_to_gt(n_span_tfrmt, .data = with_big_n_data)

If we also want big N's on col 1,2, col4, and mycol3, we will need Add 3 more rows to our ARD. One that goes down to the span1 level of specificity and two that go all the way down to my_col level of specificity. (Note: because we are keeping the parm values the same we can use the same tfrmt)

with_more_big_n_data <- tribble(
  ~group,     ~label,        ~span2,  ~span1,     ~my_col,    ~parm,   ~val,
    NA,         NA,  "column cols",  "cols 1,2",   NA  ,      "bigN",    12,
    NA,         NA,  "column cols",  "col 4"   ,   "col4"  ,  "bigN",    6,
    NA,         NA,             NA,          NA,   "mycol3",  "bigN",    6
) %>% 
  bind_rows(with_big_n_data)

print_to_gt(n_span_tfrmt, .data = with_more_big_n_data)

Page-Level Big Ns

The page_plan allows users to split up large tables in many ways. Let's suppose our page_plan splits our table so that each subtable represents a different subpopulation defined by the group variable in the ARD.

In this case, our page_plan looks something like this:

page_structure(group_val = ".default")

Because our subtables contain different populations, we want the big Ns displayed in the headers to be table-specific. To achieve this, we can supply multiple sets of big Ns in the ARD. We just need to specify the specific values of the group parameter that each big N belongs to. This functionality can be enabled via the by_page argument in big_n_structure.

For example, let's take the following data, where we have calculated big Ns for each level of the grouping variable grp:

dat <- crossing(
  grp = c("A","B"),
  lbl = c("a","b"),
  col = c("Placebo","Treatment"),
  param = "mean"
) %>% 
  mutate(val = c(1.254, 3.483, 5.123, 4.239,4.364, 8.435, 7.645, 2.312))

big_ns <- tibble(col = c("Placebo","Placebo","Treatment","Treatment"),
                 grp = c("A", "B", "A", "B"),
                 param = "bigN",
                 val = c(34, 36, 42, 39))


dat <- bind_rows(dat, big_ns)
tail(dat)

Now that we have all big Ns represented in the ARD, we can make sure of the by_page argument in the big_n_structure to ensure the various big Ns are assigned to the relevant pages defined by the page_plan.

gts <- tfrmt(
  group = grp,
  label = lbl,
  column = col, 
  param = param,
  value = val,
  body_plan = body_plan(
    frmt_structure(group_val = ".default", label_val = ".default", frmt("x.x"))
  ),
  page_plan = page_plan(
    page_structure(group_val = ".default")
  ),
  big_n = big_n_structure(param_val = "bigN", n_frmt = frmt("\nN = xx"), by_page = TRUE)
) %>% 
  print_to_gt(dat)

gts %>% gt::grp_pull(1)

gts %>% gt::grp_pull(2)

## Final note

As a final note, column names for the big N's should match the ARD column names, rather than any future names. Regardless of any renaming that happens in the `col_plan`, the column name for the big N's should be consistent with column values in the ARD. The big N's will still be added if names do get updated in the `col_plan`.  
```r
n_span_tfrmt %>% 
  tfrmt(col_plan = col_plan(
      group,
      label,
      starts_with("col"),
      new_col_3 = mycol3,
      -mycol5
    )) %>% 
  print_to_gt(.data = with_more_big_n_data)

big_n_structures provide a straightforward way of adding big N's from the ARD into the column labels. The functionality allows for big N's at any level of column spanning depending on how it is parameterized in the ARD.

GSK-Biostatistics/tlang documentation built on Dec. 11, 2024, 11:16 a.m.