26  Exercise 7: Tables, Listings and Figures

26.1 Goal

Extend the demographics table from the TLF session with an age-group row, then attempt the AE summary table the session’s own citation calls for, and find out, on real data, exactly what a missing MedDRA coding costs a submission-ready report.

26.2 Setup

Work from the project root. All three inputs are persisted:

library(tidyverse)
library(knitr)

adsl <- read_csv("data/adam/adsl.csv", show_col_types = FALSE)
adae <- read_csv("data/adam/adae.csv", show_col_types = FALSE)
adlb <- read_csv("data/adam/adlb.csv", show_col_types = FALSE)

26.3 Task

1. (Guided) An age-group row. Add an AGEGR1 row to the demographics table from Example 1, showing n (%) by arm the same way SEX and RACE do.

   Two subjects per arm have no AGEGR1, because they have no AGE (the D1 defect from The DM Domain). Decide how those four subjects should appear in your table (as a stated Missing row, folded silently into one of the age bands, or dropped) and justify the choice in one sentence. There is a wrong answer among those three; say which one and why.

2. (The trap) The table the session’s own citation calls for. OCCDS v1.1 §4 opens Example 1 by citing ICH E3 §12.2.2 and §14.3.1 for “the basic summary of adverse event frequencies,” displayed by System Organ Class and Preferred Term. Build that table for GLPX-1.

   You will not be able to. Find out precisely why, using the data rather than assuming it. Check what columns actually exist in ADAE, and what is populated in them. State the reason in one sentence, citing back to the constraint established in ADAE.

   Now build the closest thing you can build: the same summary at the level of the verbatim AETERM instead of the coded preferred term. Look at your result closely, several clinical concepts appear more than once, under slightly different spellings or capitalisation. Count how many distinct AETERM values there are, and how many remain after folding case. Then answer: why would a real submission reject a table built this way, even though every number in it is arithmetically correct?

3. (Semi-guided) The tabular twin of the figure, and one more listing. The session’s figure shows the full HbA1c trajectory. A study report also needs the single-timepoint version: a table of mean (SD) change from baseline in HbA1c at Week 26, by arm, with n.

   Build it. Then, separately, build a listing of the four subjects with missing AGE: one row each, with USUBJID, TRT01P, SEX and AGEGR1.

   Answer in one sentence each:

   • What does the Week 26 table tell a reader that the full-trajectory figure does not, and vice versa?
   • Why is a listing, not a table, the right artifact for the four missing-age subjects?

26.4 Acceptance criteria

• AGEGR1 row added with a stated, justified treatment of the four missing-age subjects
• The SOC/PT table is attempted, found impossible, and the reason is stated precisely (which column, and why it is empty)
• The AETERM-level alternative is built, with the raw and case-folded distinct-value counts both reported
• A one-sentence explanation of why the AETERM table is not an acceptable substitute, grounded in what was actually observed in the data (not just “it’s not coded”)
• Week 26 table built with correct n, mean and SD per arm
• Missing-age listing built, with both follow-up questions answered

TipHint

Task 1: reuse the pattern the session used for SEX and RACE - count(), join the arm N, format, pivot_wider(). For the missing subjects, think about what a null percentage would even mean: 2 out of what denominator?

Task 2: start by checking names(adae) for the column the SOC display would need, before writing any summarising code. ADAE already told you which coded-term columns exist and which don’t, and why. For the AETERM table, group and count exactly as the session did for AOCCFL, just keyed on AETERM instead. The fragmentation you are looking for is visible the moment you arrange() the result alphabetically.

Task 3: the Week 26 table is one filter() and one summarise() on adlb, not a new derivation: every value it needs already exists on each row. The listing is a filter() and a select() on adsl, no count() involved.

Solutions are in the paid tier.