How do I interpret findings from MR?

Results from MR analysis:

  • nsnp: the number of SNPs included in the analysis;
  • b (beta): the value of the causal estimate (i.e., slope coefficient, the causal effect of the exposure on the outcome);
  • se: standard error of the causal estimate;
  • pval: the p-value associated with the estimate.

Results from heterogeneity tests:

  • Q: Cochran’s Q statistic - to test the variability in the causal estimates generated for each SNP, which is to see the consistency in the causal estimates across all SNPs. Low heterogeneity means that MR estimates are more reliable.
  • Q_df: Cochran’s Q statistic degree of freedom;
  • Q_pval: associated p-value with Cochran’s Q statistic.

Results from directional horizontal pleiotropy test:
Horizontal pleiotropy refers to when the SNPs affect the outcome through a pathway other than exposure, which violates the MR assumption.

  • egger_intercept: to calculate the extent of average horizontal pleiotropy;
  • se: standard error associated with the intercept;
  • pval: p-value associated with the intercept.

Results visualization:

  • Forest plot: compares the causal effect calculated using the methods that include all the SNPs to the causal effect calculated using each SNP separately.
  • Scatter plot: shows the relationships between SNP effects on exposure against the SNP effects on the outcome. A scatter plot is generated for each exposure-outcome test. The different method has different line, with the slope indicating the causal association.
  • Funnel plot: a funnel plot’s asymmetry can be used to determine how reliable a certain MR method is. Wider spread implies greater heterogeneity, which may be due to horizontal pleiotropy.
  • Leave-one-out sensitivity analysis: to determine whether a single SNP is having a disproportionately larger impact on an association. Each dot represents the MR analysis excluding that specific SNP using IVW. The analysis that includes all SNPs is also displayed for comparison.

Davies, Neil M et al. BMJ (Clinical research ed.) vol. 362 k601. 12 Jul. (2018)
Skrivankova, Veronika W et al. BMJ (Clinical research ed.) vol. 375 n2233. 26 Oct. (2021)
Burgess, Stephen et al. Wellcome open research vol. 4 186. 28 Apr. (2020)