Background: Moderator effect assessment is important in personalized medicine. We mathematically prove that the average summary value is actually nonlinearly to logRR, and we assess the bias from linear meta-regression on logRR via simulation. Methods: In the meta-analysis of randomized controlled trials, the moderator effect is generally evaluated by the linear meta-regression of the logarithmic risk ratio (RR) versus the average summary value of the entire study population. Conclusions: We recommend using linear meta-regression on logarithmic odds ratio (logOR) since it has been shown that the average summary value is actually linear to logOR.