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

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

library(tfrmt)
library(dplyr)
library(gt)
library(tidyr)

A core design element of tfrmt is to offer the ability to layer tfrmts together. This ability provides an opportunity for organizations to build template table formats that can be shared, improved, and reused across multiple projects. Through layering, the users can customize the templates to create a specific table.

A template is a function that, given a set of inputs (or none), creates a standard tfrmt. We will go through their applications below.
We will create an Adverse Events table template in this vignette to demonstrate the ideas behind developing a template tfrmt.

Creating a Standard tfrmt

The first step to creating an Adverse Events table template is to determine the constants in the format. This requires there to be some standards that can be expected that allows the construction of the basic tfrmt:

First, we can provide instructions on cell formatting via the body_plan. We know that the cells should be consistently formatted as follows:

• Total counts should be whole numbers
• Cells within each treatment group column should be a combination of a count and percent of that population. When percent is 100 or zero, value should be blank.
• p-values should be formatted where: >.99 is displayed as ">0.99", <0.001 is displayed as "<0.001", and any values between should be displayed to three decimal places. Missing p-values should be replaced with "--".

Next, because our data is standardized as an analysis results dataset, we can pre-fill the expected column names. This also ensures that the data abides to the standard and names are set consistently:

• AEBODYSYS defines the body systems and overall grouping of the data.
• AETERM defines the labels to display in the table.
• value defines the values to present in the table.
• treatment defines the treatment for each value. This defines the column span.
• col defines the column for each value.
• param is the column that describes the params of the value column. "AEs" indicates that the number in value represents the total counts, "n" represents the count for the specific AE, "pct" is the percent of the treatment population, and finally "pval" is the p-value.

With this information we can construct a template AE tfrmt:

ae_tfrmt_template <- tfrmt(
    group = AEBODSYS,
    label = AETERM,
    param = param,
    column = c(treatment, col),
    value = value,
  body_plan = body_plan(
    ## All entries where the param column is `AEs` (representing total counts)
    frmt_structure(
      group_val = ".default",
      label_val = ".default",
      AEs = frmt("[XXX]")
      ),

    ## Combine entries where param column is `n` and `pct` to create a cell for
    ## that population
    frmt_structure(
      group_val = ".default",
      label_val = ".default",
      frmt_combine(
        "{n} {pct}",
        n = frmt("XXX"),
        pct = frmt_when(
          "==100" ~ "",
          "==0" ~ "",
          TRUE ~ frmt("(xx.x %)")
          )
        )
      ),

    ## All entries where param column is `pval`, format conditionally.
    ## When the value is missing, replace the NA with "--".
    frmt_structure(
       group_val = ".default", 
       label_val = ".default",
       pval = frmt_when(
         ">0.99" ~ ">0.99",
         "<0.001" ~ "<0.001",
         TRUE ~ frmt("x.xxx", missing = "--")
         )
    )
  )
)

Functionalising the Template

Now that we have our tfrmt, we can wrap it into a function to form a template. This allows us to have as many layers as we'd like. For example, we can have an organization-level template to be used across all tables, a domain-level template to be used across all tables in a given domain, and a project-level template to be used for study-specific tables.

tfrmt offers the function layer_tfrmt(), which provides the ability for layering tfrmt together. The first two arguments are the tfrmt objects to be layered. By default the body_plans of the tfrmt are joined together. We will leverage this function in the creation of our template.

See below for an example of creating a template based on a function for layering tfrmt together. In this scenario the base AE template only has body_plan values for handling total counts, case counts, and percents, but not p-values. We create another template for p-values, and then finally layer it with the study specific tfrmt.

ae_base_tfrmt_template <- function(tfrmt_obj){

  ae_base <- tfrmt(
    group = AEBODSYS,
    label = AETERM,
    param = param,
    column = c(treatment, col),
    value = value,
    body_plan = body_plan(
      ## All entries where the param column is `AEs` (representing total counts)
      frmt_structure(
        group_val = ".default",
        label_val = ".default",
        AEs = frmt("[XXX]")
        ),

      ## Combine entries where param column is `n` and `pct` to create a cell for
      ## that population
      frmt_structure(
        group_val = ".default",
        label_val = ".default",
        frmt_combine(
          "{n} {pct}",
          n = frmt("XXX"),
          pct = frmt_when(
            "==100" ~ "",
            "==0" ~ "",
            TRUE ~ frmt("(xx.x %)")
            )
          )
        )
    )
  ) 

  layer_tfrmt(x = tfrmt_obj, y = ae_base)

}

ae_pval_tfrmt_template <- function(tfrmt_obj){

  ae_pval_template <- tfrmt(
    body_plan = body_plan(
      ## All entries where param column is `pval`, format conditionally.
      ## When the value is missing, replace the NA with "--".
      frmt_structure(
         group_val = ".default", 
         label_val = ".default",
         pval = frmt_when(
           ">0.99" ~ ">0.99",
           "<0.001" ~ "<0.001",
           TRUE ~ frmt("x.xxx", missing = "--")
           )
      )
    )
  )

  layer_tfrmt(tfrmt_obj, ae_pval_template)

}

Using these templates and what we learned from the layering vignette, we can apply multiple templates cleanly within a pipe:

study_ae_tfrmt_multi_layer <- ae_base_tfrmt_template() %>% 
  ae_pval_tfrmt_template() %>% 
  tfrmt(
    title = "Adverse Events for CDISC Pilot Study",
    subtitle = "Data subset to AEs with >10% prevalence in the High Dose group",

    ## Sorting columns of rows
    sorting_cols = c(ord1, ord2),

    ## Nest Preferred terms under SOC
    row_grp_plan = row_grp_plan(label_loc = element_row_grp_loc(location = "indented")),

    ## alisgnment of columns
    col_style_plan = col_style_plan(
      col_style_structure(align = c(".",","," "), col = vars(starts_with("p_")))
    ),

    ## remove order columns from final table
    col_plan = col_plan(
      span_structure(
        treatment = c(
          "Xanomeline High Dose (N=84)" = `Xanomeline High Dose`,
          "Xanomeline Low Dose (N=84)" = `Xanomeline Low Dose`,
          "Placebo (N=86)" = Placebo
        ),
        col = c(
          `n (%)` = `n_pct` ,
          `[AEs]` = `AEs`
          )
      ),
      span_structure(
        treatment = c(
          "Fisher's Exact p-values" = fisher_pval
        ),
        col = c(
          `Placebo vs. Low Dose` = `p_low` ,
          `Placebo vs. High Dose` = `p_high`
        )
      ),
      - starts_with("ord")
    )
  )

See how this results in the same table as above:

## filter to keep only AEs with >10% prevalence in the High Dose group
data_ae2 <- data_ae %>% 
  group_by(AEBODSYS, AETERM) %>% 
  mutate(pct_high = value[col2=="Xanomeline High Dose" & param=="pct"]) %>% 
  ungroup %>% 
  filter(pct_high >10) %>% 
  select(-pct_high) %>% 
  rename(
    treatment = col2,
    col = col1
  )

study_ae_tfrmt_multi_layer %>% 
  print_to_gt(data_ae2) %>% 
  tab_options(
    container.width = 1000
  )

Alternatives to Full Templates

In addition to defining full tfrmt templates, users can also create reusable frmt templates. The benefit of this is that teams can use these frmts in their tfrmt directly and can be sure their table will comply with the value presentation expected of their organization.

Some common pre-defined frmts may include p-value displays, integers, n (%), among other things.

# defined frmts

int_frmt <- frmt("[XXX]")

pval_frmt <- frmt_when(
           ">0.99" ~ ">0.99",
           "<0.001" ~ "<0.001",
           TRUE ~ frmt("x.xxx", missing = "--")
           )

n_pct_frmt <- frmt_combine(
          "{n} {pct}",
          n = frmt("XXX"),
          pct = frmt_when(
            "==100" ~ "",
            "==0" ~ "",
            TRUE ~ frmt("(xx.x %)")
            )
          )

## frmts as functions

int_frmt_func <- function(ints = 2){
  str_exp <- paste0("[", paste0(rep("X", ints), collapse = ""), "]")
  frmt(str_exp)
}

Now that there are defined frmts, we can use them in our tfrmt to generate our table too.

tfrmt(
    group = AEBODSYS,
    label = AETERM,
    param = param,
    column = c(treatment,col),
    value = value,

    body_plan = body_plan(
      frmt_structure(
        group_val = ".default",
        label_val = ".default",
        AEs = int_frmt_func(3)
        ),

      frmt_structure(
        group_val = ".default",
        label_val = ".default",
        n_pct_frmt
        ),

      frmt_structure(
         group_val = ".default", 
         label_val = ".default",
         pval = pval_frmt
      )
    ),

    ## remove order columns from final table
    col_plan = col_plan(
      span_structure(
        treatment = c(
          "Xanomeline High Dose (N=84)" = `Xanomeline High Dose`,
          "Xanomeline Low Dose (N=84)" = `Xanomeline Low Dose`,
          "Placebo (N=86)" = Placebo
        ),
        col = c(
          `n (%)` = `n_pct` ,
          `[AEs]` = `AEs`
          )
      ),
      span_structure(
        treatment = c(
          "Fisher's Exact p-values" = fisher_pval
        ),
        col = c(
          `Placebo vs. Low Dose` = `p_low` ,
          `Placebo vs. High Dose` = `p_high`
        )
      ),
      - starts_with("ord")
    )

  ) %>% 
  print_to_gt(data_ae2) %>% 
  tab_options(
    container.width = 1000
  )

Best Practices

Here we list some of the ideas about best practices when it comes to defining tfrmt templates for use across organization. This comes from experience writing and using tfrmt and has the potential to change.

A template tfrmt is intended to provide an interface for creating complex tfrmts more easily. When creating the function, it is helpful to differentiate the aspects of the display that will remain constant from those which are likely to change across tables or studies. These changes can be captured as arguments in the function. These functions should serve to provide a way to interact with and create tfrmts for standard tables in such a way that customization for a specific table's tfrmt is simplified.

However, as with normal function development, this should be balanced with limiting the number of arguments necessary to pass and get a functional tfrmt. For example, tfrmt_sigdig accepts a minimal set of arguments, including a data.frame or tibble, to define a complex body_plan. Simplifying the way users may enter this complex set of instructions without having to manually type it is part of its strength.

In addition to minimizing the arguments to a tfrmt template, it is generally simpler to define and enforce standards around things people can see in a table. For example, the grouping, labels, and value formats are easy to see and understand. Standardizing around meta information that impacts a table but may not as easily be seen, such as value context (params), is more difficult.

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