https://orcid.org/0000-0001-7312-7078
The Mendelian randomization (MR) approach uses genetic variants as instrumental variables to study the effect of an exposure on an outcome.1 In this way, by selecting genetic variants that serve as instrumental variables for drug effects, it can also be possible to study corresponding drug side-effects and repurposing potential.2 Such an approach has recently been applied to antihypertensive drugs by two separate research groups independently of each other.3,4 Here, the authors of these two papers collaboratively discuss a strategy that can be used for validating instruments for such study.
Both of the discussed research papers studied antihypertensive drugs using genetic variants located at the locus of the gene corresponding to their respective protein targets. The work by Gill et al. selected instruments for antihypertensive drug classes as genetic variants at the corresponding protein target’s gene, promotor or enhancer region that were also associated with systolic blood pressure (SBP) at genome-wide significance (P < 5 x 10−8).3 In contrast, Walker et al.4 selected instruments as genetic variants at the corresponding protein target’s gene that were marked as the ‘best single-nucleotide polymorphism’ for relation to expression of that gene in any tissue within the Genotype-Tissue Expression (GTEx) project data.5 These genetic variants were then retained for the main analysis, regardless of the tissue that they were identified in, if there was evidence that they also had an effect on SBP in two-sample MR.4
Randomized controlled trials (RCTs) represent the gold-standard for investigating drug effects, and although they can be limited by time and resource requirements, they continue to be regarded as the definitive study design for guiding clinical practice.6 Where RCT data are available on drug effects, these can be used to explore the validity of instruments selected to study the corresponding drug in MR analyses.3 When using such a strategy, it is however important to appreciate that MR and RCTs are intrinsically different approaches, and therefore the estimates that they generate are not equivalent or interchangeable. Given that MR measures the lifelong cumulative effect of genetic predisposition, it follows that its estimates may be greater than those obtained from clinical intervention at a discrete time point, such as in a RCT. Similarly, the population characteristics for those considered within an MR and RCT setting may not coincide. Other potential risks of such comparison include the scenario where the MR estimates are biased due to incorporation of pleiotropic variants. Of note, bias in MR related to pleiotropy can also vary depending on the particular exposure–outcome pair under study—variants that affect one outcome through effects independent of the exposure may not necessarily do so for another. Despite these limitations, where MR analysis is being performed to investigate the clinical effects of a drug, it should generally follow that the findings are at least concordant with those obtained in corresponding RCTs.
In the paper by Walker et al., genetic instruments for antihypertensive drug classes were identified to explore their potential for repurposing in the prevention of Alzheimer’s disease.4 For the angiotensin-converting enzyme (ACE) inhibitor, beta-adrenoceptor blocker, calcium channel blocker and thiazide-like diuretic drug classes, corresponding RCT meta-analysis estimates are available for their effect (against placebo) on risk of coronary heart disease (CHD) and stroke.7 We can compare MR estimates with corresponding RCT meta-analysis results by using the genetic instruments that Walker et al. identified for each drug class,4 and perform two-sample inverse-variance weighted (or ratio method where only a single instrument variant is available) MR analyses for risk of CHD and stroke respectively.7 Here, we do this using publicly available genome-wide association study summary data on 60 801 CHD cases and 123 504 controls (multi-ethnic),8 and 40 585 stroke cases and 406 111 controls (European ancestry).9 MR estimates are scaled to the effect on SBP observed in RCT meta-analyses for the respective drug class (21.1 mmHg decrease for ACE inhibitors, 9.5 mmHg decrease for beta-adrenoceptor blockers, 8.9 mmHg decrease for calcium channel blockers and 12.6 mmHg decrease for thiazide-like diuretics)7 to allow comparison with RCT results, with the further assumption that odds ratio estimates approximate to relative risk estimates for CHD and stroke.3 The MR and RCT estimates are compared in Figure 1, and their concordance supports that the instruments incorporated for MR are valid.
Mendelian randomization (MR) estimates of antihypertensive drug effects on coronary heart disease and stroke risk, compared with results from randomized controlled trial (RCT) meta-analyses; the 95% confidence intervals are provided in brackets and a log10 scale is used.
Previous RCTs investigating the effect of treatment with antihypertensive medications have focused on dementia generally, rather than specifically considering Alzheimer’s disease.10,11 Furthermore, these trials studied combinations of antihypertensive medication classes. Although both the RCTs and the MR analyses performed by Walker et al. have focused on clinically diagnosed cases, thus introducing the possibility of misclassification, the different exposure and outcome definitions still preclude direct comparison of their results from being used to explore MR instrument validity.
To summarize, RCTs provide gold-standard evidence that can be used to validate putative genetic proxies for specific drug targets. Although there are limitations to such an approach, any discordance between RCT and MR findings could be used to highlight the inclusion of invalid instruments.
Data availability
The code and data used to perform the analyses presented in this article are provided on GitHub: https://github.com/venexia/rct-instrument-comparison.
Funding
This work was supported by the Wellcome 4i Clinical PhD Program at Imperial College London, the Perros Trust, and the Integrative Epidemiology Unit at the University of Bristol. The Integrative Epidemiology Unit is supported by the Medical Research Council (grant number MC_UU_00011/1, MC_UU_00011/3).
Conflict of interest: V.M.W. is currently working on a manuscript in collaboration with GlaxoSmithKline plc that explores whether Mendelian randomization can predict drug success, but she does not receive financial support from the company. N.M.D. has worked on unrelated projects funded as part of the Global Research Awards for Nicotine Dependence, which is an independent grant giving body funded by Pfizer. The remaining authors have no conflicts of interest to declare.
References
GTEx Consortium, Laboratory Data Analysis Coordinating Center Analysis Working Group, Statistical Methods Analysis Working Group, Enhancing GTEx Groups, NIH Common Fund, Biospecimen Collection Source Site Ndri, Biospecimen Collection Source Site Rpci, Biospecimen Core Resource Vari, Brain Bank Repository-University of Miami Brain Endowment, Leidos Biomedical-Project Management, Elsi Study, Genome Browser Data, Integration, Visualization EBI, Genome Browser Data Integration, Visualization-Ucsc Genomics Institute, University of California Santa Cruz Lead analysts, Laboratory Data Analysis, Coordinating Center, NIH program management, Biospecimen Collection Pathology, eQTL Manuscript Working Group,
