Sensitivity Analysis on Confounding in Observational Studies

Sensitivity Analysis in Observational Research: Introducing the E-Value

Ann Intern Med 2017. doi:10.7326/M16-2607

Confounding (or residual confounding) on unmeasured or uncontrolled variables has long been a "bête noire" of observational trialists. By definition, a confounding variable is shared by both the exposure (e.g. treatment) and the outcome (e.g. disease events) but does not lie on the causal pathway between them. Discussion sections of observational study reports often include acknowledgments of confounding (and bias) as potential limitations.

Through recent consulting work, I've become acquainted with the "E-value," which can help to quantify and/or account for confounding. According to Tyler J. VanderWeele, PhD, and Peng Ding, PhD:

"The E-value is defined as the minimum strength of association, on the risk ratio

scale, that an unmeasured confounder would need to have with

both the treatment and the outcome to fully explain away a specific

treatment–outcome association, conditional on the measured

covariates. A large E-value implies that considerable unmeasured

confounding would be needed to explain away an

effect estimate. A small E-value implies little unmeasured confounding

would be needed to explain away an effect estimate."

For much more on observational studies, including definitions and examples of >50 forms of bias, see my textbook on medical writing: https://www.amazon.com/Writing-High-Quality-Medical-Publications-Manual/dp/1498765955