jeffreyad/rpharma-2026-workshop
Developer tools이 페이지는 AI로 자동 번역되었습니다. 기술적인 세부 사항은 원문을 확인하세요. 원문 보기
2026 R/Pharma workshop: Hands-On AI Skills for Clinical Programming & Biostatistics
개요
A hands-on workshop, to be delivered at , on using Claude Code skills to accelerate two common clinical-programming tasks: 1. (ADSL, ADAE, ADVS/ADLB) from SDTM with the {admiral} R package. 2. for survival endpoints — boundaries, sample size, multiplicity, and simulation-based verification. https://jeffreyad.github.io/rpharma-2026-workshop/ - What an agent skill is (a SKILL.md workflow plus supporting references and scripts) and how Claude Code picks one up. - How the admiral skills encode function-selection logic and CDISC conventions so the agent writes QC-ready, submission-traceable R — with # REVIEW: markers at every protocol-specific decision. - How to go from a trial scenario to a fully specified group sequential design and a Word report, with the design verified by lrstat simulation. - How to run these skills yourself: install once, then drive them by describing the task in natural language. You will clone this repo and run the skills live during the session.
README
원문 README의 일부입니다. 원문 보기.
R/Pharma 2026 Workshop — AI Skills for Clinical Trial Design & ADaM Derivation
A hands-on workshop, to be delivered at R/Pharma 2026, on using Claude Code skills to accelerate two common clinical-programming tasks:
- Deriving ADaM analysis datasets (ADSL, ADAE, ADVS/ADLB) from SDTM with the {admiral} R package.
- Designing group sequential trials for survival endpoints — boundaries, sample size, multiplicity, and simulation-based verification.
Slides:
What you’ll learn
- What an agent skill is (a
SKILL.mdworkflow plus supporting references and scripts) and how Claude Code picks one up. - How the admiral skills encode function-selection logic and CDISC conventions so
the agent writes QC-ready, submission-traceable R — with
# REVIEW:markers at every protocol-specific decision. - How to go from a trial scenario to a fully specified group sequential design
and a Word report, with the design verified by
lrstatsimulation. - How to run these skills yourself: install once, then drive them by describing the task in natural language.
You will clone this repo and run the skills live during the session.
Included skills
| Skill | One-line description |
|---|---|
skills/admiral |
Parent skill — shared admiral conventions: library setup, native-pipe / exprs() style, date-imputation rules, "Y"/NA flag convention, # REVIEW: annotation and stopifnot() QC patterns. |
skills/admiral/admiral-adsl |
Derive the ADaM Subject-Level Analysis Dataset (ADSL) from SDTM DM/EX/DS — treatment dates, disposition, demographics, and population flags. |
skills/admiral/admiral-adae |
Derive the ADaM Adverse Events dataset (ADAE) from SDTM AE + ADSL — treatment-emergent flag, severity, seriousness, causality, and study-day variables. |
skills/admiral/admiral-bds |
Derive ADaM BDS findings datasets (ADVS, ADLB) — parameter assignment, baseline flagging, change from baseline, visit windowing, and ADLB normal-range derivations. |
skills/group-sequential-design |
Design group sequential trials for survival endpoints (OS, PFS, DFS): spending-function boundaries, N-first sample sizing, graphical multiplicity, NPH sensitivity, lrstat verification, and a template-driven Word report. |
Setup
1. Clone
git clone https://github.com/jeffreyad/rpharma-2026-workshop.git
cd rpharma-2026-workshop
2. Install the dependencies
Use R >= 4.2 (the skills were exercised on R 4.4.1) and, for the group-sequential-design report step, Python >= 3.8.
Rscript install_packages.R # the 12 R packages
python3 -m pip install --user python-docx # Word-report dependency
3. Verify
Rscript check_setup.R
추천 도구
다른 키워드를 입력하거나 필터를 제거해 보세요.
설치
npx skillfish add jeffreyad/rpharma-2026-workshop