There was another example where the HEIDI-outlier approach detected strong effects due to pleiotropy. The effect of LDL-c on Alzheimer’s disease (AD) was highly significant without HEIDI-outlier filtering (OR = 1.35 and PGSMR = 7.8 × 10−16) (Fig. 4). The HEIDI-outlier analysis flagged 16 SNPs, 12 of which are located in the APOE gene region (LD r2 among these SNPs < 0.05) and all of which had highly significant effects on both LDL-c and AD. Excluding these SNPs makes a more conservative GSMR test because if there is a true causal relationship of increased LDL-c with AD, then the GSMR test should remain significant based on evidence from other LDL-c associated SNPs. In fact, after removing the 16 pleiotropic SNPs, the estimated effect of LDL-c on AD was not significant (OR = 1.03, PGSMR = 0.47). Nevertheless, the multiple pleiotropic signals clustered at the APOE locus are worth further investigation (Supplementary Fig. 11).
