metaprep

Meta-analyzing effect estimates from multi-arm trials requires accounting for the dependence that arises when several treatment arms are compared to the same control group. The methods are well understood and metafor::rma.mv() implements them: you pass a block-diagonal variance-covariance matrix that encodes the dependence. The annoying part is building that matrix and keeping it synchronized with the estimates as you filter, group, and subset.

metaprep handles the bookkeeping. It provides an estimates_vcov object that keeps coefficient estimates and their variance-covariance matrix in sync through dplyr operations, plus thin wrappers around metafor that read from it directly.

Installation

Install the development version from GitHub:

# install.packages("remotes")
remotes::install_github("acoppock/metaprep")

Example

Prepare a fit from each trial, bind them together, and build a synchronized estimates_vcov object:

library(dplyr)
library(metaprep)

set.seed(343)
fit_1 <- lm(Y ~ Z, data = data.frame(Y = rnorm(50),  Z = sample(c("T0", "T1"), 50, TRUE)))
fit_2 <- lm(Y ~ Z, data = data.frame(Y = rnorm(100), Z = sample(c("T0", "T1", "T2"), 100, TRUE)))

fits_df <- bind_rows(
  `Study 1` = prep_fit(fit_1, term = "ZT1"),
  `Study 2` = prep_fit(fit_2, term = c("ZT1", "ZT2")),
  .id = "study"
)

ev <- as_estimates_vcov(fits_df)
ev
#> <estimates_vcov>
#> # 3 estimates with 3x3 vcov matrix
#> 
#> # A tibble: 3 × 7
#>   id    study   term  estimate std.error statistic p.value
#>   <chr> <chr>   <chr>    <dbl>     <dbl>     <dbl>   <dbl>
#> 1 1     Study 1 ZT1    -0.153      0.233    -0.654   0.516
#> 2 2     Study 2 ZT1     0.134      0.228     0.588   0.558
#> 3 3     Study 2 ZT2     0.0943     0.225     0.419   0.676

dplyr verbs keep the vcov matrix in sync with the estimates:

ev |> filter(study == "Study 2")
#> <estimates_vcov>
#> # 2 estimates with 2x2 vcov matrix
#> 
#> # A tibble: 2 × 7
#>   id    study   term  estimate std.error statistic p.value
#>   <chr> <chr>   <chr>    <dbl>     <dbl>     <dbl>   <dbl>
#> 1 2     Study 2 ZT1     0.134      0.228     0.588   0.558
#> 2 3     Study 2 ZT2     0.0943     0.225     0.419   0.676

Meta-analyze with the metafor wrapper, which pulls the estimates and vcov from the object automatically:

library(metafor)
#> Loading required package: metadat
#> Loading required package: numDeriv
#> 
#> Loading the 'metafor' package (version 5.0-1). For an
#> introduction to the package please type: help(metafor)
ev |> rma_mv_helper(yi = estimate, random = ~ 1 | id)
#> 
#> Multivariate Meta-Analysis Model (k = 3; method: REML)
#> 
#> Variance Components:
#> 
#>             estim    sqrt  nlvls  fixed  factor 
#> sigma^2    0.0000  0.0000      3     no      id 
#> 
#> Test for Heterogeneity:
#> Q(df = 2) = 0.7915, p-val = 0.6732
#> 
#> Model Results:
#> 
#> estimate      se    zval    pval    ci.lb   ci.ub    
#>   0.0027  0.1505  0.0178  0.9858  -0.2923  0.2977    
#> 
#> ---
#> Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

See vignette("metaprep") for the full workflow, including component extraction, grouped meta-analyses, and what happens under the hood.