Evidence-appraisal glossary
E-value
The E-value is the minimum strength of association, on the risk-ratio scale, that an unmeasured confounder would need with both the exposure and the outcome to fully explain away an observed association, beyond the variables already adjusted for. Larger E-values signal findings more robust to hidden confounding.
Also called: E value, VanderWeele-Ding E-value.
What it is
Observational studies can never adjust for every confounder, so a real question is: how strong would some unmeasured factor have to be to erase the result? The E-value, proposed by VanderWeele and Ding (2017), answers exactly that. It is the minimum association (risk-ratio scale) an unmeasured confounder would need with both the exposure and the outcome, on top of measured covariates, to reduce the observed estimate to no effect.
How to use it when reading a study
- Two E-values are usually reported: one for the point estimate, one for the confidence-interval limit nearest the null. The second is the more honest stress test.
- Compare the E-value to associations of known, measured confounders. If a plausible unmeasured factor could easily be that strong, the finding is fragile.
- A large E-value does not prove causation; it only shows the result is hard to explain away.
- E-values near 1 mean minimal confounding could overturn the claim.
This is a plain-language methodology definition for reading research. It is general education, not medical advice.