Mendelian Randomization and Infection: Pitfalls and Promises

Abstract

Mendelian randomization (MR) is an increasingly common study design in infectious diseases (ID). It holds promise for identifying causes and consequences of infections where conventional epidemiology has struggled, and can highlight plausible drug targets, as shown in successful coronavirus disease 2019 (COVID-19) trials (baricitinib, tocilizumab). However, many current applications provide limited insight due to violations of core assumptions, yielding uninterpretable results. This article reviews MR principles, assumptions, and specific challenges in ID. We highlight examples violating key assumptions, noting that MR studies using infection as an exposure are particularly prone to bias compared to using infection as an outcome. We discuss the future of MR in ID, emphasizing appropriate application to address causal questions unanswerable by other methods and capitalize on emerging opportunities where MR can provide unique insights.

Keywords: Mendelian randomization; causal inference; epidemiology; genetics; malaria.

Figure 1.

A directed acyclic graph describing Mendelian randomization. An illustrative example might be the use of genetic variants associated with iron levels as instruments to explore the causal effect of iron status on pneumonia risk. The 3 IV assumptions are also visualized: it is possible to test whether the instruments influence the exposure, but it is not possible to falsify whether the dashed arrows (IV2 and IV3) exist, although sensitivity analyses do exist. Abbreviations: IV, instrumental variable; SNP, single nucleotide polymorphism.

Figure 2.

An example of a Mendelian randomization scatter plot. On the x-axis we plot the effect of the instrument on the exposure (in this case CRP), and on the y-axis we plot the effect on the outcome (pneumonia). The plotted lines represent meta-analyses of the individual SNPs, and the slope of each line represents the causal effect estimate. In the below figure, SNPs that increase CRP tend to also increase the odds of pneumonia, and there is therefore evidence of an effect of increasing CRP on the odds of pneumonia. Abbreviations: CRP, C-reactive protein; SNP, single nucleotide polymorphism. Error bars represent standard errors.

Figure 3.

A noncomprehensive schematic of infection progression and potentially measurable outcomes in GWAS. The blue boxes represent outcomes of infection, all of which are potentially measurable in GWAS and themselves influenced by genetics. The orange box (pathogen presence) is independent of genotype, the green boxes represent plausible genetic influences that would not be measured as outcomes in infection GWAS but could clearly influence progression through the stages. This is clearly a simplification, and there is likely circularity in the diagram. However, the key feature is that when measuring, for example, hospitalization with infection, the person must have progressed through the stages before, and the genetic influences on each stage are highly likely to be different. Abbreviation: GWAS, genome-wide association study.