Skip to main content
Fig. 2 | BMC Medicine

Fig. 2

From: Sex differences in systemic metabolites at four life stages: cohort study with repeated metabolomics

Fig. 2

Illustration of the potential for collider bias in analyses of biological sex and metabolite levels when adjusting for BMI. a Estimates of the total effect of biological sex (male vs female: exposure of interest) on metabolite levels at each time point (shown here is 8 years for example) are not prone to confounding by BMI, socioeconomic, or behavioural factors because whilst such factors could influence metabolite levels, they could not plausibly influence biological sex assignment at conception and would be considered mediators of effect. b Conditioning on BMI (potential mediator) measured at the time of metabolite assessment could produce estimates of a direct effect of sex on metabolite levels at each time point, but this could also induce non-causal associations between sex and metabolite levels due to an induced association between sex and socioeconomic or behavioural factors (confounders of the mediator-outcome associations). Such bias may be correctable via adjustment for confounders of the mediator-outcome association, but these could be time-varying and unmeasured. We considered the potential for collider bias to be low here given the strong agreement between effect estimates of BMI with these same metabolites from several observational and Mendelian randomisation analyses (i.e. low potential for mediator-outcome confounding) [29,30,31,32]

Back to article page