Episensr: an R package for deterministic and probabilistic bias analysis of epidemiologic data

Abstract

Random and systematic errors are present in any observational studies. If random error can be reduced by increasing the sample size, systematic error (or bias) is due to issues with study design or the methods used to obtain the study data and is not influenced by sample size. Bias can result from classification errors, selection bias, and unmeasured or unknown confounders. Bias analysis tries to quantify the direction, magnitude, and uncertainty of the bias affecting an estimate of association [Rothman et al., 2012, Lash et al., 2009]. This assessment is seldom realized due in part to the lack of appropriate and easy-to-use tools. The R package episensr allows to perform a quantitative bias analysis, adjusting the relative risk for selection bias, unmeasured confounding, and unior multi-dimensional misclassification of the exposure, outcome, or covariates. Both deterministic and probabilistic sensitivity analyses are implemented, with a choice of probability distributions for the bias parameters. episensr—R episens—Stata sensmac—SAS Excel