23 Exercise 4: ADSL

23.1 Goal

Extend the ADSL from the ADSL session with a second age grouping and an analysis age, then construct the case the GLPX-1 data never gives you: population flags that legitimately disagree.

23.2 Setup

Work from the project root. The finished SDTM is the input:

library(tidyverse)
library(sdtm.terminology)

dm <- read_csv("data/sdtm/dm.csv", show_col_types = FALSE,
               col_types = cols(.default = col_character()))

Build the ADSL spine as in Example 1 before starting, or read the persisted copy from data/adam/adsl.csv.

23.3 Task

(Guided) A second age grouping. Add AGEGR2 and AGEGR2N using three bands - <50, 50-64, >=65. Confirm the one-to-one relationship holds and that the two variables are null on exactly the same rows. Why is there no is_term() check to run on AGEGR2?
(The trap) Analysis age. Four subjects have a null AGE because their birth dates were incomplete. Suppose the SAP specifies an imputation for age-subgroup analyses: a missing day becomes the 15th of the month, and a missing month and day become 30 June. Derive AAGE implementing that rule, leaving AGE untouched.

Note how many of the four subjects the rule actually rescues, and what happens to the one it does not.

Two things to get right, and both are the point of the task: say which ADaMIG rule stops you writing the imputed value into AGE itself, and check whether the ADaMIG’s condition for including AAGE is actually met here.
(Semi-guided) Flags that disagree. In GLPX-1 every randomized subject was dosed, so RANDFL, ITTFL and SAFFL are all "Y" for all 400. Take subject GLPX1-101-010, construct the scenario where they were randomized but never received a dose, and show what each of the three flags becomes.

Then answer in one sentence each: which population would that subject be counted in for an efficacy table, which for a safety table, and why a null flag would be wrong for either.

23.4 Acceptance criteria

AGEGR2/AGEGR2N one-to-one, null on the same rows, and the no-codelist reasoning stated
AAGE derived under the SAP rule with AGE unmodified; the governing rule named and AAGE’s conditional requirement checked
The randomized-not-dosed scenario produces RANDFL = "Y", ITTFL = "Y", SAFFL = "N", with no nulls
Efficacy vs safety population membership explained

Hint

Task 1: the same structural rule you used for AGEGR1/AGEGR1N in the session applies (ADaMIG v1.3, §3.2). For the second part, recall what made PARAMCD uncheckable in ADaM Concepts.

Task 2: look up AAGE in ADaMIG v1.3 §3.2. Read its Core designation and the exact wording of when it is required. The rule preventing you from writing into AGE is in §3.1.1, and the session quotes it.

Task 3: you are changing one input, not one flag. Ask which condition each flag is derived from - SAFFL’s condition is about exposure, the other two are about randomization, then let the derivations you already wrote produce the three values themselves.

Solutions are in the paid tier.