Meta Analysis / 医学Meta分析

Meta-Analysis / 医学Meta分析

R-based comprehensive meta-analysis. Every module ships reproducible R code.

Language / 语言

• This skill is English-default; auto-switches to Chinese when OS locale is zh-*.
• Detailed language policy → references/language_policy.md.
• Chinese user guide → README_zh-CN.md
• English user guide → README.md

Triage (ct-base §5.2)

On first user message, classify into one of three:

If unsure between Simple and Complex → give short reply + optional expansion hint.

Traceability / Grounding (ct-base §5.1)

All factual/assertive claims must cite source: specific ref-*.md section (e.g., §3.6) or official guideline. If a claim has no verifiable source → mark ⚠️ official verify and ask user to confirm against official text.

Initialization

1. Integrity check: R source code is embedded as Python templates (scripts/r_*.py + r_templates.py) to comply with marketplace restrictions on .R files. On first use, run:

   bash scripts/check_integrity.sh   # exit 0 = ready
   

   • If scripts/*.R exists → ready; if missing → script auto-runs python r_templates.py to generate from templates.
2. R env: Rscript --version → missing → https://cran.r-project.org/
3. Install (manual): This skill does NOT auto-install R packages. Run bash scripts/check_integrity.sh or pre-analysis self-check to see missing packages; then manually install in R, or Rscript scripts/setup_packages.R to view full list.
   • Packages: metafor meta dmetar netmeta ggplot2 esc clubSandwich robumeta bayesmeta multinma gemtc dosresmeta survmeta mada metagear gridExtra ggforestplot svglite
   • ⚠️ Manual install connects to CRAN; use trusted network.
4. Workspace: Create meta_analysis/ + output/ in current directory (⚠️ will write files).
5. Memory: read ~/.workbuddy/MEMORY.md for R config (only R-related config keys; unrelated personal content is ignored and never sent anywhere).

Interactive Guide

Triage path: Vague → Level 1 menu (7 categories). Select → Level 2 with data-format hints. Sufficient info → skip menu, run analysis directly.

=== Level 1: Main ===
1️⃣ Pairwise Meta     4️⃣ Effect Size      7️⃣ Review Workflow
2️⃣ Heterogeneity & Bias  5️⃣ Visualization
3️⃣ Advanced Models   6️⃣ Study Quality

=== Level 2: Sub-Menus (excerpt) ===
[1] Pairwise: Binary(OR/RR/RD) | Continuous(SMD/MD) | Pre-calc(yi+CI) | Survival(HR/IRR) | Correlation(r→Zr)
[2] Heterogeneity: I²/Q/τ² | Subgroup | Meta-regression | Pub Bias(Egger/Begg/Trim-fill) | Sensitivity | GOSH | Baujat
[3] Advanced: Multi-arm NMA(netmeta) | Bayesian NMA Stan(multinma) | JAGS(gemtc) | Multilevel | Multivariate | IPD | Dose-Resp(dosresmeta) | Survival(survmeta) | TSA | Bootstrap
[4] Effect Size: Mean→d | t/F/r→d | d↔Hedges'g | d↔logOR | r↔Z | OR↔logOR | Batch
[5] Viz: Forest(5 themes) | Funnel | Bubble | GOSH | Network | League Table | RoB Traffic-light | Power Curve | Drapery
[6] Quality: RoB 1.0/2.0 | ROBINS-I | GRADE | PRISMA Checklist
[7] Workflow: PRISMA Flow | Screening GUI | PDF Batch-download (⚠️ opt-in) | Digitize | NNT Meta

Data formats & full details → references/interactive_menu.md

Other formats? Install @skill:statdata-transfer for 50+ format conversion.

Core Functions

Reusable API (Mandatory)

Any analysis MUST call existing functions — never rewrite the full pipeline inline. Full function list + calling examples → references/advanced_api.md.

Security & Scope

Execution model: R analysis runs locally; skill does NOT auto-install R packages (missing → list + prompt manual install). PDF full-text download from external services ONLY on explicit user instruction. Analysis artifacts written to meta_analysis/ + output/ by default.

Not clinical judgment: Results require professional interpretation.

No literature DB search: Does not search literature databases; only downloads full text when user provides DOI/PMID.

Output

analysis_complete.R + forest/funnel (.svg+.png) + results_summary.md + data_backup.csv.

Output figures as editable SVG; editing + journal format conversion → references/svg_editing.md.

Units

See references/units.md for the full 5-field schema (input / output / dependency / AI autonomy / consumer units).

References

Project Files

README.md | README_zh-CN.md | CHANGELOG.md | AGENTS.md | LICENSE (MIT © 2025 medstatstar) | requirements.txt | assets/icon.svg

Changelog

Version / fix log → CHANGELOG.md.