Meta Analysis

Comprehensive R-based meta-analysis skill. Covers all RevMan 5.x features + Stata metareg/mvmeta equivalents + esc effect size conversions + clubSandwich/robumeta cluster-robust variance estimation. Supports random/fixed-effect models, multilevel/multivariate meta, network meta, Bayesian meta, power analysis. All analysis ships reproducible R code. / 基于 R 的全方位 Meta 分析技能，覆盖 RevMan 全部功能 + Stata metareg/mvmeta 等价实现 + esc 效应量转换 + clubSandwich/robumeta 聚类稳健方差估计，支持随机/固定效应模型、多水平/多变量元分析、网络Meta、贝叶斯Meta、功效分析，所有分析提供可复现 R 代码。

Wintone Zhang@medstatstar

Install

openclaw skills install @medstatstar/meta-analysis

Meta-Analysis Skill / Meta 分析技能

R-based comprehensive meta-analysis. All RevMan 5.x features + Stata metareg/mvmeta equivalents + esc conversions + cluster-robust RVE. Every analysis ships reproducible R code. 基于 R 的全方位 Meta 分析。覆盖 RevMan 全部功能 + Stata 等价 + esc 转换 + 聚类稳健方差估计。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, and software tools such as R and RevMan already provide robust statistical engines. However, for most clinical professionals, these tools carry a significant learning curve — users must either master statistical programming or rely on dedicated statisticians to complete an analysis. This skill is designed to lower that barrier entirely. With natural-language dialogue, any clinical professional can independently conduct a complete meta-analysis, producing publication-ready results backed by professional-grade R code. Every step is powered by R's specialized meta-analysis packages. The complete, reproducible R code is provided alongside every result, enabling users to verify, re-run, and build confidence in the accuracy of their findings.

Meta分析是医学领域中至关重要的循证医学分析技术，现有软件（如 R、RevMan 等）均可实现完善的 Meta 分析。但对于临床医生而言，这些工具均存在一定的使用门槛——需要掌握相应编程方法，或依赖专业统计人员才能完成分析。本技能的目的是让临床医学工作者通过自然语言的提示词对话方式，顺畅自如地独立完成 Meta 分析，并输出完全符合出版要求的分析结果，从而大幅提升其科研能力。整个技能完全基于 R 统计环境中的专业软件包实现 Meta 分析，且 100% 提供相应的 R 运行代码，供用户检查和重跑，确保结果准确无误。

Initialization / 初始化

R env / 检查R: Rscript --version → if missing, direct to https://cran.r-project.org/
Install Strategy / 安装策略:
text
```
1️⃣ Install all now / 一次性全部安装（推荐）  2️⃣ Install on demand / 按需安装
```
- If 1: install metafor meta dmetar netmeta ggplot2 esc clubSandwich robumeta bayesmeta gridExtra ggforestplot via install.packages(); dmetar from GitHub if needed
- If 2: skip, install when triggered
Workspace / 工作目录: create meta_analysis/ + output/
Memory / 记忆: read ~/.workbuddy/MEMORY.md for R config

Data Input / 数据输入

Step A: Choose type / 选择类型

text

1 Binary (events + n) → OR/RR/RD    二分类（事件数+n）
2 Continuous (mean±SD+n) → SMD/MD   连续型（均值±SD+n）
3 Pre-calculated (yi + 95%CI)       已有效应量+CI
4 Upload CSV/Excel                   上传文件

Other formats (SPSS/Stata/SAS/JSON...)? Install @skill:statdata-transfer to convert 50+ formats. 其他格式（SPSS/Stata/SAS/JSON...）？ 安装 @skill:statdata-transfer 可转换 50+ 格式。 Required variable frameworks per data type / 各类型所需变量框架:

Binary: study, n_exp, event_exp, n_ctrl, event_ctrl[, year] → 二分类

Continuous: study, n_exp, mean_exp, sd_exp, n_ctrl, mean_ctrl, sd_ctrl[, year] → 连续型

Effect sizes: study, effect_type, effect_size, lower95, upper95[, year] → 效应量

Step B: Templates / 数据模板

See references/data_templates.md for full examples per type. 详见 references/data_templates.md。

Step C: Validation / 验证

Check	Pass	Fail
Records ≥ 2	✅ X studies	⚠️ Need ≥ 2
Required columns	✅ Found	❌ Missing: XXX
No missing values	✅ Complete	⚠️ X missing
Valid numbers	✅ OK	⚠️ Outliers

Confirm → reply "1" to continue. 确认后回复"1"继续。

Core Functions / 核心功能

Module	R Trigger / 触发词
Effect Size / 效应量	`escalc()`, `esc_mean_sd()` — SMD, OR, RR, RD, HR, ROM, ZCOR
Pooling / 合并	`rma()`, `metabin()` — FE, RE(DL/HK), MH, Peto
Forest Plot / 森林图	`forest()` + ggplot2 (5 themes: minimal/lancet/jama/revman/custom)
Heterogeneity / 异质性	I², Q, τ², 95% PI — auto-reported
Publication Bias / 偏倚	`funnel()`, `regtest()`, `ranktest()`, `trimfill()`, `selmodel()`
Subgroup / 亚组	`rma(mods = ~ factor - 1)` + QB test
Meta-Regression / 元回归	`rma(yi, vi, mods)` + bubble plot
Sensitivity / 敏感性	Leave-one-out, GOSH, quality/sample-size filter
Network Meta / 网络meta	`netmeta::netmeta()` — node-splitting, SUCRA, league table
Bayesian / 贝叶斯	`bayesmeta::bayesmeta()` — half-normal prior, MCMC diagnostics
Power / 功效	`dmetar::power.analysis()`
Risk of Bias / 偏倚风险	RoB 1.0/2.0 templates + traffic-light plot

RevMan 1:1 Coverage / RevMan 对应

Full mapping in references/revman_complete.md. 详细代码见 references/revman_complete.md。

RevMan	R	✅
Dichotomous / 二分类	`metabin()`, `rma()`	✅
Continuous / 连续型	`metacont()`, `rma()`	✅
Forest/Funnel / 森林/漏斗	ggplot2 + `funnel()`	✅
Subgroup / 亚组	`rma(mods = ~ factor - 1)`	✅
Heterogeneity / 异质性	I²/Q/τ² auto	✅
Sensitivity / 敏感性	Leave-one-out, GOSH	✅
Pub. Bias / 发表偏倚	Egger/Begg/Trim-fill	✅
GRADE / RoB	Structured + templates	✅
Network / 网络	`netmeta` wrapper	✅
Bayesian / 贝叶斯	`bayesmeta`	✅

Stata Equivalents / Stata 等价

Details in references/stata_to_r_mapping.md. 详见 references/stata_to_r_mapping.md。

Stata	R
`metareg`	`rma(yi, vi, mods, method="REML")` + `permutest(fit, iter=1000)`
`mvmeta` UN/CS/HCS/AR1/ID/DIAG	`rma.mv(yi, V, random=~outcome\|study, struct="UN"/"CS"/"HCS"/"AR1"/"ID"/"DIAG")`
V matrix (multi-arm)	`build_V_matrix_CS()`: Cov(dᵢ,dⱼ) = 1/n₀ + dᵢ·dⱼ/(2·N)

Effect Size Conversion / 效应量转换

Details in references/esc_robust_meta.md. 详见 references/esc_robust_meta.md。

Conversion	R
Mean/SD/n → Cohen's d / Hedges' g	`esc::esc_mean_sd()` + J = 1 − 3/(4df − 1)
d ↔ logOR	logOR = d·π/√3
r ↔ Fisher's z	z = 0.5·ln((1+r)/(1−r))
OR/CI → logOR+SE	`esc::esc_or()`
Batch convert	`metafor::escalc()`

Cluster-Robust RVE / 聚类稳健方差估计

For dependent effect sizes (multi-outcome/multi-arm). Details in references/esc_robust_meta.md. 处理依赖效应量。详见 references/esc_robust_meta.md。

Package	Use Case
`robumeta::robu()`	≥10 studies, multiple outcomes (ρ=0.8, small=TRUE)
`clubSandwich::vcovCR(type="CR2")`	Small-sample correction
`metafor::vcalc()` + `impute.vcov()`	Manual V matrix

Workflow: dependency check → V matrix → fit (≥10: robu; <10: CR2) → robust CI + I². 流程：依赖检测 → V矩阵 → 拟合（≥10: robu；<10: CR2）→ 稳健CI + I²。

Advanced / 高级方法

Multilevel / IPD / Bayesian / Dose-Response / GOSH / Power → references/advanced_analysis.md 多水平/IPD/贝叶斯/剂量反应/GOSH/功效 → references/advanced_analysis.md

Security & Scope / 安全与范围

Local only / 仅本地: all R runs locally, no data transmitted
Auto-install / 自动安装: user-level library, no system changes
Not a substitute for clinical judgment / 不替代临床判断
No database searching / 不执行数据库检索

Output / 输出

analysis_complete.R + forest/funnel plots (.svg+.png) + results_summary.md + data_backup.csv. All SVG — editable in Illustrator/Inkscape. 所有图形为矢量 SVG，可在 Illustrator/Inkscape 中编辑。

Citations / 引用

R, metafor, meta, dmetar, netmeta, esc, robumeta, clubSandwich — full list in references/citations.md.

References / 参考

File	Content
`r_packages.md`	Package details & install / 包详解与安装
`revman_complete.md`	RevMan→R 1:1 code / RevMan→R 1:1 代码
`stata_to_r_mapping.md`	Stata→R mapping / Stata→R 映射
`esc_robust_meta.md`	esc + RVE reference / esc + 稳健方差估计
`advanced_analysis.md`	Multilevel/IPD/Bayesian / 多水平/IPD/贝叶斯
`data_templates.md`	Data input templates per type / 各类型数据输入模板
`citations.md`	Full citations / 完整引用