Understanding `tern` functions

In tern: Create Common TLGs Used in Clinical Trials

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

Understanding tern functions

Every function in the tern package is designed to have a certain structure that can cooperate well with every user's need, while maintaining a consistent and predictable behavior. This document will guide you through an example function in the package, explaining the purpose of many of its building blocks and how they can be used.

As we recently worked on it we will consider summarize_change() as an example. This function is used to calculate the change from a baseline value for a given variable. A realistic example can be found in LBT03 from the TLG-catalog.

summarize_change() is the main function that is available to the user. You can find lists of these functions in ?tern::analyze_functions. All of these are build around rtables::analyze() function, which is the core analysis function in rtables. All these wrapper functions call specific analysis functions (always written as a_*) that are meant to handle the statistic functions (always written as s_*) and format the results with the rtables::in_row() function. We can summarize this structure as follows:

summarize_change() (1)-> a_change_from_baseline() (2)-> [s_change_from_baseline() + rtables::in_row()]

The main questions that may arise are:

1. Handling of NA.
2. Handling of formats.
3. Additional statistics.

Data set and library loading.

library(dplyr)
library(tern)

## Fabricate dataset
dta_test <- data.frame(
  USUBJID = rep(1:6, each = 3),
  AVISIT = rep(paste0("V", 1:3), 6),
  ARM = rep(LETTERS[1:3], rep(6, 3)),
  AVAL = c(9:1, rep(NA, 9))
) %>%
  mutate(ABLFLL = AVISIT == "V1") %>%
  group_by(USUBJID) %>%
  mutate(
    BLVAL = AVAL[ABLFLL],
    CHG = AVAL - BLVAL
  ) %>%
  ungroup()

Classic use of summarize_change().

fix_layout <- basic_table() %>%
  split_cols_by("ARM") %>%
  split_rows_by("AVISIT")

# Dealing with NAs: na_rm = TRUE
fix_layout %>%
  summarize_change("CHG", variables = list(value = "AVAL", baseline_flag = "ABLFLL")) %>%
  build_table(dta_test) %>%
  print()

# Dealing with NAs: na_rm = FALSE
fix_layout %>%
  summarize_change("CHG", variables = list(value = "AVAL", baseline_flag = "ABLFLL"), na_rm = FALSE) %>%
  build_table(dta_test) %>%
  print()

# changing the NA string (it is done on all levels)
fix_layout %>%
  summarize_change("CHG", variables = list(value = "AVAL", baseline_flag = "ABLFLL"), na_str = "my_na") %>%
  build_table(dta_test) %>%
  print()

.formats, .labels, and .indent_mods depend on the names of .stats. Here is how you can change the default formatting.

# changing n count format and label and indentation
fix_layout %>%
  summarize_change("CHG",
    variables = list(value = "AVAL", baseline_flag = "ABLFLL"),
    .stats = c("n", "mean"), # reducing the number of stats for visual appreciation
    .formats = c(n = "xx.xx"),
    .labels = c(n = "NnNn"),
    .indent_mods = c(n = 5), na_str = "nA"
  ) %>%
  build_table(dta_test) %>%
  print()

What if I want something special for the format?

# changing n count format and label and indentation
fix_layout %>%
  summarize_change("CHG",
    variables = list(value = "AVAL", baseline_flag = "ABLFLL"),
    .stats = c("n", "mean"), # reducing the number of stats for visual appreciation
    .formats = c(n = function(x, ...) as.character(x * 100))
  ) %>% # Note you need ...!!!
  build_table(dta_test) %>%
  print()

Adding a custom statistic (and custom format):

# changing n count format and label and indentation
fix_layout %>%
  summarize_change(
    "CHG",
    variables = list(value = "AVAL", baseline_flag = "ABLFLL"),
    .stats = c("n", "my_stat" = function(df, ...) {
      a <- mean(df$AVAL, na.rm = TRUE)
      b <- list(...)$.N_row # It has access at all `?rtables::additional_fun_params`
      a / b
    }),
    .formats = c("my_stat" = function(x, ...) sprintf("%.2f", x))
  ) %>%
  build_table(dta_test)

For Developers

In all of these layers there are specific parameters that need to be available, and, while rtables has multiple way to handle formatting and NA values, we had to decide how to correctly handle these and additional extra arguments. We follow the following scheme:

Level 1: summarize_change(): all parameters without a starting dot .* are used or added to extra_args. Specifically, here we solve NA values by using inclNAs = TRUE always in rtables::analyze(). This will keep NA values to the analysis function a_*. Please follow the way na_rm is used in summarize_change, and you will see how to retrieve it from ... only when you need it. In this case, only at the summary() level. na_str, instead is set only on the top level (in the rtables::analyze() call). We may want to be statistic-dependent in the future, but we still need to think how to accomplish that. We add the rtables::additional_fun_params to the analysis function so to make them available as ... in the next level. Note that they all can be retrieved with list(...)[["na_rm"]].

Level 2: a_change_from_baseline(): all parameters starting with a dot . are ideally used or transmitted into lower functions from here. Mainly .stats, .formats, .labels, and .indent_mods are used only at this level. We also bring forward extra_afun_params to the ... list for the statistical function. Notice the handling for additional parameters in the do.call() function.

Level 3 and beyond: s_* functions. In this case s_summary is at the end used and the result brought into the main a_* function.

tern documentation built on April 4, 2025, 12:10 a.m.