25  Exercise 6: ADLB

25.1 Goal

Extend the ADLB from the ADLB session with a categorised change, then build the two things GLPX-1’s single baseline definition never forces you to build: a derived baseline that requires DTYPE, and a second baseline definition that requires BASETYPE.

25.2 Setup

Work from the project root. Both inputs are persisted:

library(tidyverse)
library(sdtm.terminology)

lb   <- read_csv("data/sdtm/lb.csv", show_col_types = FALSE,
                 col_types = cols(.default = col_character()))
adlb <- read_csv("data/adam/adlb.csv", show_col_types = FALSE)
adsl <- read_csv("data/adam/adsl.csv", show_col_types = FALSE)

Build the ADLB as in the session before starting, or read the persisted copy from data/adam/adlb.csv.

25.3 Task

1. (Guided) Categorised change. For PARAMCD == "HBA1C" at AVISIT == "WEEK 26", add CHGCAT1 grouping CHG into three bands: "Reduction >= 1%", "Reduction 0 to <1%", and "No reduction". Add its numeric counterpart CHGCAT1N.

   Look up CHGCATy in ADaMIG v1.3 §3.3.4.1 before you start and note its Core designation. Then answer: which structural rule ties CHGCAT1N to CHGCAT1, and is CHGCAT1 checkable against a CDISC codelist?

   Cross-tabulate the result by TRTA. The clinically meaningful number should be obvious.

2. (The trap) A derived baseline. Suppose the SAP is amended: for HbA1c only, baseline becomes the average of the screening and the last pre-dose value, rather than the single last pre-dose value.

   Re-derive ABLFL, BASE and CHG for HbA1c under this definition. Getting the arithmetic right is the easy part; the exercise is about what else must change.

   Two requirements, and both are the point:

   • The averaged baseline is a new record, not an edit to an existing one. Add it as a row, and set DTYPE on it, choosing a value the ADaMIG itself gives as an example.
   • State what AVISIT should be on that row, and why the answer is not obvious.
   • Report how many subjects actually have two pre-dose HbA1c values. Not all of them do, and what you decide to do about the rest is part of the answer.

   Then recompute the week-26 treatment difference and compare it with the session’s -1.116. Report the number you get, whatever it is.

3. (Semi-guided) Two baselines at once. Now suppose the study report needs both definitions side by side: the session’s last-pre-dose baseline and task 2’s averaged baseline, for the same parameter, in the same dataset.

   Restructure so both coexist, and answer in one sentence each:

   • Which variable becomes required, and what exactly does it have to be populated on?
   • Why can you not simply add a second column called BASE2?
   • How many rows does HbA1c now have per subject, and what does that do to the “one row per subject per parameter per timepoint” description of BDS?

25.4 Acceptance criteria

• CHGCAT1/CHGCAT1N one-to-one, null on the same rows, with the no-codelist reasoning stated
• The averaged baseline exists as a new row with DTYPE populated, not as a modified original record
• The AVISIT decision for the derived row is stated and justified
• The recomputed week-26 difference is reported as obtained
• BASETYPE identified, with the scope of “must be non-null” stated precisely
• All three follow-up questions in task 3 answered

TipHint

Task 1: the structural rule for a category and its numeric counterpart is the one AGEGR1/AGEGR1N followed in ADSL and SEVGR1 followed in Exercise 5. For the codelist question, ask what kind of variable CHGCATy is rather than searching for a codelist name.

Task 2: DTYPE is in ADaMIG v1.3 §3.3.5, which lists example values directly; one of them is exactly this case. Note that sdtm.terminology cannot help you here, because it packages SDTM controlled terminology and DTYPE is an ADaM codelist. Knowing which terminology a variable belongs to is part of the skill. For AVISIT, re-read the three precisions quoted in the session’s Timing section, especially the one about what AVISIT does not mean. There is no single right answer, but there is a wrong one: silently reusing BASELINE when a row belongs to no visit.

Task 3: ADaMIG v1.3 §4.2.1.6, Rule 6 is the worked example for exactly this situation. Read it before restructuring. The answer to “why not BASE2” is in what BDS does with additional analysis concepts: it adds rows, not columns.

Solutions are in the paid tier.