A general (mathematical) relationship between an observed effect estimate and the confounding that could account for it, applicable to any observational hazard/risk ratio — including the small egg–CVD and egg–T2D cohort estimates. It converts “residual confounding could explain this away” from a qualitative worry into a quantitative threshold: the E-value states how strong an unmeasured both-ways confounder (e.g. an unadjusted healthy-user factor) would have to be to reduce a given association to the null. The closed-form value and its application are given in the attached argument.