exposure) does not correlate with the effect the instrument has on the outcome (i.e. the InSIDE assumption: Instrument Strength Independent of Direct Effect). This is a weaker assumption than the assumption of no pleiotropy. MR-Egger may, however, be biased when the NOME (NO Measurement Error) assumption is violated – i.e. the assumption that the SNP-exposure associations are known rather than estimated. Violation of NOME can be quantified with the I2 statistic, which ranges between 0 and 1. A value below 0.9 indicates a considerable risk of bias. This bias can be corrected for with MR-Egger simulation extrapolation SIMEX70. Since I2 ranged between 0.7-0.9 for our analyses, we report results from MR-Egger SIMEX in Table 3. The Weighted Mode methods can produce an unbiased result, as long as the most common causal effect estimate is a consistent estimate of the true causal effect: the Zero Modal Pleiotropy Assumption (ZEMPA)69. Finally, we performed GSMR, a method which leverages power from multiple genetic variants while accounting for LD between these variants64. Because GSMR accounts for LD, genetic variants that were included in GSMR instruments were pruned at a higher threshold of R2<0.05 (instead of R2<0.001 for the other MR analyses). Zhu et al.
