Table 3.

Comparison of MR estimates for LDL‐cholesterol on coronary artery disease between GCTA‐GSMR and MR‐PRESSO

Method	Settings	Raw estimate			Outlier adjusted				Runtime, sec	Additional comments
		Causal estimate	SE	p value	Causal estimate	SE	p value	SNPs filtered
GCTA‐GSMR	LD‐matrix precomputed	NA	NA	NA	0.432	0.022	4.15E‐85	9	2.5	Runtime was on the basis of the analysis portion only. The computation of the LD‐matrix needs to be done via GCTA separately.
MR‐PRESSO	Nb = 1000, outlier p val = 0.05	0.402	0.071	5.02E‐08	0.462	0.027	5.70E‐36	48	39.1	Outlier test unstable with only 1000 simulations to compute the null distribution (i.e. cannot obtain pval of outlier < 0.188).
	Nb = 10,000, outlier p val = 0.05	0.402	0.071	5.02E‐08	0.408	0.028	2.56E‐30	45	375.2
	Nb = 50,000, outlier p val = 0.05	0.402	0.071	5.02E‐08	0.408	0.028	2.56E‐30	45	1871.2

Abbreviations: GCTA‐GSMR, genome‐wide complex trait analysis‐generalized summary mendelian randomization; MR‐PRESSO, mendelian randomization pleiotropy residual sum and outlier; SNP, single nucleotide polymorphism.

Causal estimate refers to the estimated effect size (log(OR) on coronary artery disease (CAD) risk per standard deviation increase in genetically predicted LDL‐cholesterol (LDL‐c). SE refers to the respective standard errors of the causal estimate. Nb denote the number of simulation replicates required to generate the null distribution used in the MR‐PRESSO outlier tests. The data for these traits were extracted from publicly available GWAS summary statistics (LDL‐c from http://csg.sph.umich.edu/willer/public/lipids2013/; CAD from http://www.cardiogramplusc4d.org/data‐downloads/).