Reading guide
Read the book in sequence or enter through the path that matches your role. Either way, reading alone is not the completion criterion: every chapter closes with exercises and a case study, and the stack only sticks when you run it.
How the book is organized
Fifteen chapters, five parts, three layers that decay at different speeds:
| Layer | What it contains | Where it appears | Half-life |
|---|---|---|---|
| L1 — The rules | CDISC traceability, risk-based validation thinking, GxP reasoning | Parts I and IV | A decade |
| L2 — The stack | admiral, metacore, rtables, cards, teal, targets, rhino | Parts II–IV | Years; habits transfer even when APIs move |
| L3 — The frontier | LLM agents, MCP servers, natural language to CDISC | Part V | Months; treat every specific claim as perishable |
The layers are taught together, the way they exist in production: no admiral chapter makes sense without the traceability rules above it, and no AI chapter makes sense without the validation wall it has to climb. Each frontier chapter carries a volatile layer note with a last-verified date — re-check tool specifics before relying on them.
Three reading paths
The SAS programmer migrating (or being migrated)
Read Part I in full, then the 4 → 6 → 7 spine — ADaM and tables. Keep Chapter 3 as ammunition for the budget meetings: it is written for the people who sign, not only the people who code.
| Stage | Chapters | Evidence of completion |
|---|---|---|
| Vocabulary and context | 1, 2, 3 | You can draw the SDTM→ADaM→TLF flow and name the R package at each station |
| The production spine | 4, 5, 6, 7 | You rebuild one study dataset and one AE table from your own shop’s spec |
| The case for switching | 3, 9 | You can argue migration cost and validation posture in front of an audit committee |
The R engineer entering clinical
Read Chapter 1 for vocabulary, then jump to Part IV. Your engineering instincts are an asset; Chapter 9 is the license to use them. The data-chain chapters will feel foreign at first — that is expected, and Chapter 4 is the gentlest entry.
| Stage | Chapters | Evidence of completion |
|---|---|---|
| Domain vocabulary | 1 | You stop saying “clean data” and start saying which standard governs the shape |
| Engineering under supervision | 9, 10, 11 | You write a qualification memo and a validation evidence file QA would accept |
| Backfill the chain | 4, 5, 7 | You can explain why a derivation brick is easier to validate than a script |
The statistician or data-science lead
Read 2, 3, 9, and Part V. You will not write the code; you will decide whether the code is allowed. The checklist tables in each chapter are designed to be screenshotted into your governance deck.
A chapter study cycle
- Skim the TL;DR. Every chapter opens with one; decide what you are looking for before the details arrive.
- Run the code. Examples are written to be executed in a fresh R session. Predict key outputs before running.
- Steal the checklist. Each chapter ends its core argument with a selection or audit checklist — adapt it to your shop’s context.
- Do the exercises. Three per chapter, moving from recall to application.
- Work the case study. The closing case study puts the chapter’s decision in a realistic setting; write down your answer before comparing it with the discussion.
- Note the verification date. Frontier chapters state when their claims were last verified. If that date is old, re-check before citing specifics.
Code, data, and honesty about execution
Code fences display teaching code; a few examples are intentionally illustrative sketches (marked as pseudocode in comments) that name a pattern rather than a runnable API. Simulated data stands in for real trial data throughout — no chapter contains patient data, and no example should be pointed at a real study without your institution’s procedures and acceptance criteria.
The companion volume
This book assumes general R fluency and spends no chapters on it. If you want the from-zero engineering foundation — functions, testing, reproducible delivery, and AI-assisted habits outside the regulated context — start with the companion volume, Modern R in Practice, and return here for the clinical stack. Together they are the curriculum the author wishes someone had handed him at the wall.