The EValue package allows users to calculate bounds and E-values forunmeasured confounding in observational studies and meta-analyses. Thepackage also includes functions for the assessment of selection bias anddifferential misclassification and the joint impact of all three typesof bias.
You can install the released version of EValue fromCRAN with:
install.packages("EValue")Then load the package:
library(EValue)E-values are simple to calculate. For example, the E-value for theassociation between cigarette smoking and lung cancer as observed byHammond 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 NAFor more on E-values for unmeasured confounding, see thevignette.
More complex assessment of several biases is also easy. To bound thebias due to unmeasured confounding, selection bias, and differentialoutcome misclassification, we can use background knowledge about thestrength 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] 2.386364Read more about how to specifymultiplebiases and see several workedexamples.
If all you need to do is calculate an E-value for unmeasuredconfounding, just try out theonline calculator.Graphical interfaces are also linked under each of the types ofsensitivity analysis in the header.