Sensitivity Analysis for Multiple Biases in Meta-Analyses

Meta-analyses can be compromised by studies' internal biases (e.g., confounding in nonrandomized studies) as well as by publication bias. This package conducts sensitivity analyses for the joint effects of these biases (per Mathur (2022) ). These sensitivity analyses address two questions: (1) For a given severity of internal bias across studies and of publication bias, how much could the results change?; and (2) For a given severity of publication bias, how severe would internal bias have to be, hypothetically, to attenuate the results to the null or by a given amount?

multibiasmeta

multibiasmeta is an R package that provides bias correction and sensitivity analysis for the joint effects of within-study and across-study biases in meta-analysis (per Mathur, 2022).

Installation

You can install multimetabias from CRAN:

install.packages("multibiasmeta")

Or you can install the development version of multibiasmeta from GitHub with:

# install.packages("devtools")
devtools::install_github("mathurlabstanford/multibiasmeta")

Example

Calculate the meta-analytic effect size estimate, correcting for 1) publication bias with an assumed selection ratio of 4 (i.e., affirmative results are 4x more likely to be published than nonaffirmative ones); 2) internal biases with assumed mean values (on the same scale as yi values).

library(multibiasmeta)
multibias_meta(yi = meta_meat$yi, vi = meta_meat$vi, biased = TRUE, selection_ratio = 4,
               bias_affirmative = log(1.5), bias_nonaffirmative = log(1.1))

Calculate how high mean internal bias would need to be to attenuate the effect size estimate to the null, assuming a selection ratio of 4.

multibias_evalue(yi = meta_meat$yi, vi = meta_meat$vi, selection_ratio = 4,
                 biased = !meta_meat$randomized)