20  Exercise 1: SDTM Demographics

20.1 Goal

Rebuild the core of the GLPX-1 DM dataset yourself, including the three decisions that make it more than a rename, and defend the result.

20.2 Setup

Work from the project root, with the session’s packages loaded:

library(tidyverse)
library(sdtm.terminology)

dm_raw <- read_csv("data/raw/dm_raw.csv", show_col_types = FALSE,
                   col_types = cols(.default = col_character()))
ex_raw <- read_csv("data/raw/ex_raw.csv", show_col_types = FALSE,
                   col_types = cols(.default = col_character()))

Do the tasks in order: each builds on the last. The DM Domain session is the reference; try each task before looking back at how the session did the equivalent step.

20.3 Task

1. (Guided) The identifier block. Reduce dm_raw to one row per subject, then build STUDYID, DOMAIN, USUBJID and SUBJID. Confirm (with code, not by inspection) that USUBJID is unique and that no identifier would change if a subject moved sites.
2. (Guided) The clock. Derive RFSTDTC, RFENDTC, RFXSTDTC and RFXENDTC from the dosing file. Then compute the study day of 2024-03-05 and 2024-03-06 for subject 101-012 (first dose 2024-03-06) and check you get −1 and 1.
3. (Semi-guided) Defend the nulls. Identify every subject whose AGE must be null, and write one sentence per subject saying why
   • naming the precision of the birth date you refused to extend. Then name two columns that must be present-but-null and two that must be absent, and say which core designation produces each fate.
4. (Open-ended) The decision log. Write the two decision-log entries a reviewer would want for this dataset: the one-row resolution for 101-012 and the SEX resolution for 111-003. Each entry: what conflicted, what rule was applied, why that rule, and what was queried. Aim for the register of study documentation, not a chat message.

20.4 Acceptance criteria

• 400 rows, USUBJID unique, identifiers built only from immutable facts
• Reference dates populated for all subjects; study-day check returns −1 and 1 (no zero)
• Four AGE nulls, each with a stated reason; Expected-vs-Permissible fates correctly assigned
• Both decision-log entries name the rule, not just the outcome
• Every CT-governed column you touched was checked with is_term(), not asserted from memory

TipHint

Task 2’s check is the two-branch rule from the session: before the reference date, the study day is the plain date difference; on or after it, the difference plus one. If you got 0 for 2024-03-05, you used one branch for both dates.

Task 4’s entries need three ingredients each: the conflict, the rule (one row per person; site-entered source outranks a downstream copy), and the action taken (SUPPDM note; query to the lab).

Solutions are in the paid tier.