Sensitivity Analyses for Unmeasured Confounding and Other Biases in Observational Studies and Meta….

The EValue R package

The EValue package allows users to calculate bounds and E-values for unmeasured confounding in observational studies and meta-analyses. The package also includes functions for the assessment of selection bias and differential misclassification and the joint impact of all three types of bias.

Installation

You can install the released version of EValue from CRAN with:

install.packages("EValue")

Then load the package:

library(EValue)

Examples

E-values are simple to calculate. For example, the E-value for the association between cigarette smoking and lung cancer as observed by Hammond and Horn in 1958:

evalues.RR(est = 10.73, lo = 8.02, hi = 14.36)
#>             point    lower upper
#> RR       10.73000  8.02000 14.36
#> E-values 20.94777 15.52336    NA

For more on E-values for unmeasured confounding, see the vignette.

More complex assessment of several biases is also easy. To bound the bias due to unmeasured confounding, selection bias, and differential outcome misclassification, we can use background knowledge about the strength of the biases to propose sensitivity analysis parameters:

biases <- multi_bias(confounding(),
                     selection("general", "increased risk"),
                     misclassification("exposure", rare_outcome = TRUE))

multi_bound(biases,
            RRUcY = 2, RRAUc = 1.5,
            RRSUsA1 = 1.25, RRUsYA1 = 2.5,
            ORYAaS = 1.75)
#> [1] 2.386364

Read more about how to specify multiple biases and see several worked examples.

Other options

If all you need to do is calculate an E-value for unmeasured confounding, just try out the online calculator. Graphical interfaces are also linked under each of the types of sensitivity analysis in the header.