Parallelize 'metafor' functions

Henrik Bengtsson

+ =

The futurize package allows you to easily turn sequential code into parallel code by piping the sequential code to the futurize() function. Easy!

TL;DR

library(futurize)
plan(multisession)
library(metafor)

dat <- escalc(measure = "RR", ai = tpos, bi = tneg,
              ci = cpos, di = cneg, data = dat.bcg)
fit <- rma(yi, vi, data = dat)
pr <- profile(fit) |> futurize()

Introduction

This vignette demonstrates how to use this approach to parallelize metafor functions such as profile(), rstudent(), cooks.distance(), and dfbetas().

The metafor package provides a comprehensive collection of functions for conducting meta-analyses in R. It supports fixed-effects, random-effects, and mixed-effects (meta-regression) models and includes functions for model diagnostics and profiling. Several of these computations involve fitting the model repeatedly, making them excellent candidates for parallelization.

Example: Likelihood profile for a random-effects model

The profile() function computes the likelihood profile for model parameters such as the variance component in a random-effects meta-analysis. For example, using the built-in BCG vaccine dataset:

library(metafor)

## Calculate log risk ratios and sampling variances
dat <- escalc(measure = "RR", ai = tpos, bi = tneg,
              ci = cpos, di = cneg, data = dat.bcg)

## Fit a random-effects model
fit <- rma(yi, vi, data = dat)

## Compute likelihood profile
pr <- profile(fit)

Here profile() is calculated sequentially. To calculate in parallel, we can pipe to futurize():

library(futurize)
library(metafor)

dat <- escalc(measure = "RR", ai = tpos, bi = tneg,
              ci = cpos, di = cneg, data = dat.bcg)
fit <- rma(yi, vi, data = dat)
pr <- profile(fit) |> futurize()

This will distribute the profile computations across the available parallel workers, given that we have set up parallel workers, e.g.

plan(multisession)

The built-in multisession backend parallelizes on your local computer and works on all operating systems. There are other parallel backends to choose from, including alternatives to parallelize locally as well as distributed across remote machines, e.g.

plan(future.mirai::mirai_multisession)

and

plan(future.batchtools::batchtools_slurm)

Supported Functions

The following metafor functions are supported by futurize():

• profile() for ‘rma.uni’, ‘rma.mv’, ‘rma.ls’, and ‘rma.uni.selmodel’
• rstudent() for ‘rma.mv’
• cooks.distance() for ‘rma.mv’
• dfbetas() for ‘rma.mv’

Without futurize: Manual PSOCK cluster setup

For comparison, here is what it takes to parallelize profile() using the parallel package directly, without futurize:

library(metafor)
library(parallel)

## Calculate log risk ratios and sampling variances
dat <- escalc(measure = "RR", ai = tpos, bi = tneg,
              ci = cpos, di = cneg, data = dat.bcg)

## Fit a random-effects model
fit <- rma(yi, vi, data = dat)

## Set up a PSOCK cluster
ncpus <- 4L
cl <- makeCluster(ncpus)

## Compute likelihood profile in parallel
pr <- profile(fit, parallel = "snow", ncpus = ncpus, cl = cl)

## Tear down the cluster
stopCluster(cl)

This requires you to manually create and manage the cluster lifecycle. If you forget to call stopCluster(), or if your code errors out before reaching it, you leak background R processes. You also have to decide upfront how many CPUs to use and what cluster type to use. Switching to another parallel backend, e.g. a Slurm cluster, would require a completely different setup. With futurize, all of this is handled for you - just pipe to futurize() and control the backend with plan().