Correction of 'Bias factor, maximum bias and the E-value'

Arvid Sjölander, Tyler J VanderWeele
2021 International Journal of Epidemiology  
We would like to correct an incorrect statement in the paper 'Bias factor, maximum bias and the E-value: insight and extended applications', recently published in the International Journal of Epidemiology by Cusson and Infante-Rivard. 1 In the box of Key Messages, the authors state: 'As currently available, the E-value cannot be used with all possible direction combinations between the unmeasured confounder and both the exposure and the outcome; new formulae we propose make this possible.' This
more » ... his possible.' This is incorrect. The E-value, as it was originally proposed (i.e. 'as currently available'), can in fact be used regardless of the direction of association between the unmeasured confounders and the exposure and the outcome. Before tracing the source of this confusion, we recap the main results of Ding and VanderWeele, 2,3 who originally introduced the E-value. In Cusson and Infante-Rivard's notation, Ding and VanderWeele defined the observed risk ratio as and the 'true' (i.e. causal) risk ratio as RR true ¼ P k pðY ¼
doi:10.1093/ije/dyab055 pmid:33779748 pmcid:PMC8407863 fatcat:bssvhqu7ubdsdmpkkkxjwv5ily