Bayesian Meta-Analysis: General Anesthesia vs Conscious Sedation During Endovascular Thrombectomy

Analysis code for a Bayesian meta-analysis comparing general anesthesia (GA) vs conscious sedation / non-GA during endovascular thrombectomy (EVT) for acute ischemic stroke, based on randomized controlled trials.

Outcomes

• Primary: mRS 0–2 at 90 days (functional independence)
• Secondary (binary): mRS 0–1, mortality at 90 days, TICI 2b–3 (successful reperfusion), symptomatic ICH, pneumonia, hypotension
• Secondary (continuous): Hospital length of stay, ICU length of stay, NIHSS change (24h – baseline)

Statistical approach

All primary analyses use Bayesian random-effects meta-analysis via the bayesmeta R package with weakly informative priors:

Parameter	Binary outcomes (OR)	Continuous outcomes (MD)
Effect prior (mu)	N(0, 1.5)	N(0, 5)
Heterogeneity prior (tau)	Half-normal(0, 0.5)	Half-normal(0, 2)

Results are reported as posterior median with 95% credible intervals (CrI) and the posterior probability of benefit for GA, P(benefit).

Repository structure

R/
  run_all.R                    # Master script (sources everything in order)
  meta_analysis.R              # Primary Bayesian meta-analyses for all outcomes
  publication_bias.R           # Harbord's test (binary) / Egger's test (continuous)
  meta_regression.R            # Bayesian meta-regression (mortality, TICI 2b-3)
  sensitivity_analyses.R       # Low RoB, anterior only, HKSJ, leave-one-out
  trial_sequential_analysis.R  # TSA for mRS 0-2
data/
  data_prepared.xlsx         # Template with column definitions (no patient data)

Scripts must be run sequentially (each depends on objects from prior scripts). Use source("R/run_all.R") to run the full pipeline.

Data format

The analysis expects a file data/data_prepared.xlsx with one row per study. The included template shows the required column structure with example values. Replace the example rows with your extracted study data.

Requirements

R >= 4.1 with the following packages:

install.packages(c(
  "readxl", "tidyverse", "writexl",
  "bayesmeta", "metafor",
  "RTSA"
))

How to run

# Set working directory to the project root
setwd("path/to/this/repo")

# Run full analysis pipeline
source("R/run_all.R")

# Or run individual scripts
source("R/meta_analysis.R")   # creates bma_* objects
source("R/publication_bias.R") # requires bma_* from meta_analysis.R

License

This code is provided for academic reproducibility. Please cite the associated publication if you use or adapt this code.