Meta-Analysis / 医学Meta分析

R-based comprehensive meta-analysis. Every module ships reproducible R code. / 基于R的全方位Meta分析，所有分析提供可复现R代码。

Purpose / 技能目的

Meta-analysis is a cornerstone of evidence-based medicine. However, existing tools carry a learning curve — users must master statistical programming or rely on statisticians. This skill lowers that barrier entirely: any clinical professional can independently conduct meta-analysis via natural-language dialogue, producing publication-ready results backed by professional R code. Every step is powered by R's specialized packages with complete reproducibility. / Meta分析是循证医学的核心技术，但现有工具使用门槛较高。本技能让临床医学工作者通过自然语言对话独立完成Meta分析，输出符合出版要求的结果，全部基于R专业软件包，100%提供可复现R代码，供用户检查和重跑。

Language Policy / 语言策略

Default to English; switch to Chinese when a Chinese-language environment is detected.
Common modules (frontmatter description/displayName, Core Functions table, triggers, etc.) ship full English + Chinese bilingual content.
Document sections that are language-agnostic use bilingual headings in English / Chinese order (e.g., ## Core Functions / 核心功能).
Complex / less commonly used modules (e.g., advanced Bayesian NMA, survmeta, dosresmeta) may temporarily be English-only.
默认英文；检测到中文环境时切换为中文提示。
常用模块（frontmatter description/displayName、Core Functions 表、触发词等）备英文 + 中文两套。
文档标题（不区分语言者）采用「英 / 中」顺序双语（如 ## Core Functions / 核心功能）。
复杂 / 少用模块（如高级贝叶斯 NMA、survmeta、dosresmeta）可暂只英文。

Initialization / 初始化

Integrity check / 完整性自检（必做，最先执行）: 本技能将 R 源码以 Python 模板内嵌（scripts/r_*.py + r_templates.py）以规避部分技能市场（如 SkillHub）不允许发布 .R 文件的限制；安装后 scripts/ 下可能暂缺 .R。请先运行自检：
```
bash scripts/check_integrity.sh   # 退出码 0=可用(已生成或原本齐全)
```
- 若 scripts/*.R 齐全 → 直接通过；
- 若缺失 → 脚本会自动执行 python r_templates.py 从内嵌模板生成全部 .R，无需用户手动下载。
Do not rewrite R code by yourself; let it auto-generate when missing. / 不要自行重写 R 代码；缺失时让它自动生成即可。
R env / 检查 R: Rscript --version → missing → https://cran.r-project.org/
Install / 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 them in R (e.g. core packages metafor, meta, netmeta), or run Rscript scripts/setup_packages.R to view the full list.
- Packages needed: metafor meta dmetar netmeta ggplot2 esc clubSandwich robumeta bayesmeta multinma gemtc dosresmeta survmeta mada metagear gridExtra ggforestplot svglite
- ⚠️ Manual installation will connect to CRAN / GitHub and write to your R library (may execute package install hooks); please use a trusted network.
Workspace: Create meta_analysis/ + output/ in the current working directory (⚠️ will write files — ensure the directory is correct and you have write permission).
Memory: read ~/.workbuddy/MEMORY.md for R config

Interactive Guide / 交互式引导

Design / 设计: Vague prompt → Level 1 menu (7 categories). Select → Level 2 with data format hints. Sufficient info → run analysis directly. / 模糊提示 → Level 1主菜单(7类) → 选择后Level 2子菜单(含数据格式) → 信息充足后直接分析。

=== Level 1: Main / 主菜单 ===
1️⃣ Pairwise Meta / 两组Meta     4️⃣ Effect Size / 效应量转换    7️⃣ Review Workflow / 系统评价流程
2️⃣ Heterogeneity & Bias / 异质性偏倚  5️⃣ Visualization / 可视化
3️⃣ Advanced Models / 高级模型     6️⃣ Study Quality / 研究质量

=== Level 2: Sub-Menus (excerpt) / 子菜单（示例）===
[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(UN/CS/AR1...) | IPD | Dose-Resp(dosresmeta) | Survival(survmeta) | TSA(run_tsa 自实现) | Bootstrap(bootmeta)
[4] Effect Size: Mean→d | t/F/r→d | d↔Hedges'g | d↔logOR | r↔Z | OR↔logOR | Batch(escalc)
[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 (⚠️ 需联网从外部服务获取全文，请确认版权/授权) | Digitize | NNT Meta

Data formats & full details → references/interactive_menu.md

Other formats? Install @skill:statdata-transfer for 50+ format conversion. / 其他格式？安装 @skill:statdata-transfer。

Core Functions / 核心功能

Module	R Packages & Functions
Single-Group Meta / 单组率均值	`metaprop()` `metamean()` `metainc()` `metacor()` `metarate()`
Pairwise Meta / 两组Meta	`metabin()` `metacont()` `metagen()` `rma()` — FE/RE(DL/HK)/MH/Peto
Effect Size / 效应量	`escalc()` `esc_mean_sd()` — SMD, OR, RR, RD, HR, ROM, ZCOR
Forest/Funnel / 森林漏斗	`forest()` `funnel()` + ggplot2 (5 themes)
Heterogeneity / 异质性	I², Q, τ², H², 95% PI — auto-reported
Publication Bias / 偏倚	`regtest()` `ranktest()` `trimfill()` `selmodel()`
Subgroup & Reg / 亚组回归	`rma(mods=~factor-1)` + `bubble()`
Sensitivity / 敏感性	Leave-one-out, Cumul, GOSH, quality filter
Bayesian Pairwise / 贝叶斯两组	`bayesmeta::bayesmeta()` — half-normal/JC prior
Bayesian NMA / 贝叶斯NMA	`run_bayes_nma_multinma()`(Stan) · `run_bayes_nma_gemtc()`(JAGS) — 封装 multinma/gemtc
Multilevel/MV Meta / 多水平多元	`rma.mv()` + UN/CS/HCS/AR1/ID/DIAG
Survival Meta / 生存Meta	`run_surv_meta()` — 封装 survmeta（HR/logHR 合并）
TSA & Diagnostics	`run_tsa()`(自实现,O'Brien-Fleming 边界) · `baujat()` `drapery()` `bootmeta()`
Dose-Response / 剂量反应	`run_dose_resp()` — 封装 dosresmeta（连续 smd / 二分类 gl）
Diagnosis Meta / 诊断准确性	`mada::reitsma()` bivariate + SROC
RVE Robust / 聚类稳健	`robumeta::robu()` `clubSandwich::vcovCR(CR2)`
Review Workflow / 评价流程	`metagear`: PRISMA, screen, PDF, digitize
Quality / 质量	`rob()` RoB 1.0/2.0/ROBINS-I + GRADE
Power / 功效	`run_power_curve()`(自实现,无依赖) + subgroup power

Reusable API / 复用接口（强制）

任何分析必须调用已有函数，禁止从零编写完整分析脚本。 完整函数清单、调用示例与重依赖封装（TSA / 剂量反应 / 生存 Meta / 贝叶斯 NMA）见 references/advanced_api.md。 Rule: never rewrite the full pipeline inline — source() the skill scripts and call the functions. Full API reference → references/advanced_api.md.

Security & Scope / 安全与范围

Execution model / 运行模型: R analysis runs on your local machine; this skill does NOT auto-install R packages (when missing, it only lists package names and prompts manual installation — you decide whether to connect to the network and install). It downloads full-text PDFs from external services ONLY on your explicit instruction (network involved — please confirm copyright/authorization yourself). Analysis artifacts are written to the meta_analysis/ and output/ directories in your current workspace by default. Not clinical judgment / 不替代临床判断: Results must be interpreted in conjunction with professional background. No literature DB search / 不检索文献库: This skill does not include literature database search; it only downloads full text on demand when you provide a DOI/PMID.

Output / 输出

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

Output figures as editable SVG; editing methods and journal format conversion (EPS/PDF/TIFF) → references/svg_editing.md. / 图形以可编辑 SVG 输出；编辑方式与期刊格式转换（EPS/PDF/TIFF）见 references/svg_editing.md。

References / 参考

File	Content
`interactive_menu.md`	Full Level 2 menus + data formats + dialogue examples
`revman_complete.md`	RevMan→R 1:1 code mapping
`stata_to_r_mapping.md`	Stata metareg/mvmeta→R equivalents
`esc_robust_meta.md`	Effect size conversions + RVE reference
`advanced_analysis.md`	Multilevel/IPD/Bayesian/Dose-Resp/Power
`single_group_meta.md`	metaprop/metamean/metainc/metacor + NNT
`bayesian_nma.md`	multinma (Stan) + gemtc (JAGS) full workflow
`survival_meta.md`	survmeta + KM pseudo-IPD reconstruction
`tsa_diagnostics.md`	TSA(run_tsa 自实现) + Baujat + Drapery + Bootmeta
`diagnosis_meta.md`	mada::reitsma bivariate + SROC
`review_workflow.md`	metagear: PRISMA, screening, PDF batch, digitize
`data_templates.md`	Data input templates per type
`citations.md`	Full citation list
`r_packages.md`	Package details & installation
`advanced_api.md`	复用接口（强制）+ 重依赖封装：TSA / 剂量反应 / 生存 / Bayesian NMA
`svg_editing.md`	SVG editing tools & journal format conversion

Meta Analysis

安装到 Tokeny（自动）

SKILL.md