Meta-Analysis Pipeline

AI-assisted, end-to-end meta-analysis with reproducible tooling.

From research question to manuscript-ready output -- powered by Claude Code.

---

Quick Start

# 1. Setup (one-time)
cp .env.example .env        # Add your API keys
cd tooling/python && uv init

# 2. Create a new project
uv run tooling/python/init_project.py --name my-meta-analysis

# 3. Edit your research question
# Open projects/my-meta-analysis/TOPIC.txt and paste your topic

# 4. Launch Claude Code and say:

"Start project my-meta-analysis" or "See projects/my-meta-analysis/TOPIC.txt and start"

That's it. Claude will handle the rest.

Don't have a topic yet? Say:

"Help me brainstorm a topic"

Claude will guide you through an interactive session to develop your research question and create the project for you.

---

What Claude Does

1. Creates your project in projects/<your-project-name>/
2. Reads your topic from projects/<your-project-name>/TOPIC.txt
3. Asks clarifying questions (databases, outcomes, dates)
4. Runs the 9-stage pipeline automatically
5. Generates manuscript-ready outputs

Pipeline Overview

All stages are created inside projects/<your-project-name>/:

Stage	Output
01_protocol	pico.yaml, eligibility.md
02_search	dedupe.bib
03_screening	decisions.csv
04_fulltext	manifest.csv
05_extraction	extraction.csv
06_analysis	figures/, tables/
07_manuscript	manuscript.pdf
08_reviews	grade_summary.md
09_qa	final_qa_report.md

---

Example: Completed Project

See a real meta-analysis → projects/ici-breast-cancer/

This is a 99% complete meta-analysis on immune checkpoint inhibitors in triple-negative breast cancer:

• 5 RCTs, N=2,402 patients
• Primary outcome: RR 1.26 (95% CI 1.16-1.37), p=0.0015, ⊕⊕⊕⊕ HIGH quality
• Manuscript: 4,921 words (Lancet Oncology compliant)
• Time invested: ~14 hours (vs 100+ hours manual)

Quick tour:

1. projects/ici-breast-cancer/README.md - Project overview
2. projects/ici-breast-cancer/00_overview/FINAL_PROJECT_SUMMARY.md - Key findings
3. projects/ici-breast-cancer/07_manuscript/ - Complete manuscript (5 sections)

Use as template for your own meta-analysis workflow.

Project Structure

meta-pipe/
├── ma-*/                    # Framework code modules
├── docs/archive/            # Archived documentation
├── tooling/                 # Shared tools and scripts
└── projects/                # All your meta-analysis projects
    ├── ici-breast-cancer/   # Example: complete meta-analysis
    ├── legacy/              # Historical data (pre-2026-02-08)
    └── your-project/        # Your new projects
        ├── 01_protocol/
        ├── 02_search/
        ├── ...
        ├── 09_qa/
        └── TOPIC.txt

Note: Projects are isolated and NOT tracked by Git (see .gitignore). Only the example project ici-breast-cancer/ is tracked for reference.

---

Documentation

Doc	Purpose
GETTING_STARTED.md	Manual step-by-step guide
API Setup	Database API keys
CLAUDE.md	Agent behavior (auto-loaded)
tooling/scripts/	Utility scripts

Citation

If you use meta-pipe in your research, please cite it:

AMA Format:

Lin HT, Yeh JT. meta-pipe: AI-assisted, end-to-end meta-analysis pipeline with reproducible tooling. GitHub; 2025. Accessed 2026. https://github.com/htlin222/meta-pipe

BibTeX:

@software{lin_metapipe_2025,
  author       = {Lin, Hsieh-Ting and Yeh, Jiunn-Tyng},
  title        = {meta-pipe: AI-assisted, end-to-end meta-analysis pipeline with reproducible tooling},
  year         = {2025},
  url          = {https://github.com/htlin222/meta-pipe},
  note         = {Accessed: 2026}
}

Requirements

• uv (Python)
• R ≥ 4.2 + renv
• cmake (required for building R packages like fs on macOS ARM)
• Quarto
• API keys in .env

---

Citation

If you use meta-pipe in your research, please cite it.

AMA format:

Lin HT, Yeh JT. meta-pipe: AI-assisted, end-to-end meta-analysis pipeline with reproducible tooling. GitHub; 2026. Accessed March 22, 2026. https://github.com/htlin222/meta-pipe

BibTeX:

@software{lin2026metapipe,
  author    = {Lin, Hsieh-Ting and Yeh, Jiunn-Tyng},
  title     = {meta-pipe: AI-Assisted, End-to-End Meta-Analysis Pipeline with Reproducible Tooling},
  year      = {2026},
  url       = {https://github.com/htlin222/meta-pipe},
  note      = {Software}
}