Sensitivity to unmeasured confounding
We calculated E-values (package Evalue [26] in R], and also bias-corrected estimates [27] (package episensr [28] in R) to account for bias arising from the fact that we did not control for the SLCO1B3 c.334T>G (rs4149117) and UGT1A9 c.98T>C (rs72551330) SNPs. For the latter purpose, we used literature data to generate meta-analytical estimates (packagemeta [29] in R) of strength of association (we used ratio of means, ROM [30], as an effect measure) between these SNPs and steady-state exposure to MPA (see ESM – Supplemental Methods C, for details).