Mendelian randomization for nephrologists

Confounding is a major limitation of observational studies. Mendelian randomization (MR) is a powerful study design that uses genetic variants as instrumental variables to enable examination of the causal effect of an exposure on an outcome in observational data. With the emergence of large-scale genome-wide association studies in nephrology over the past decade, MR has become a popular method to establish causal inferences. However, MR is a complex and challenging methodology that requires careful consideration to ensure robust results. This review article aims to summarize the basic concepts of MR, its application and relevance in nephrology, and the methodological challenges and limitations as well as discuss the current guidelines for design and reporting. With reference to a clinically relevant example of examining the causal relationship between the estimated glomerular filtration rate and cancer, this review outlines the key steps to conducting an MR study, including the key considerations and potential pitfalls at each step. These include defining the clinical question, selecting the data sources, identifying and refining appropriate genetic variants by considering linkage disequilibrium and associations with potential confounders, harmonization of variants across data sets, validation of the genetic instrument by assessing its strength, estimation of the causal effects, confirming the validity of the findings, and interpreting and reporting results. Confounding is a major limitation of observational studies. Mendelian randomization (MR) is a powerful study design that uses genetic variants as instrumental variables to enable examination of the causal effect of an exposure on an outcome in observational data. With the emergence of large-scale genome-wide association studies in nephrology over the past decade, MR has become a popular method to establish causal inferences. However, MR is a complex and challenging methodology that requires careful consideration to ensure robust results. This review article aims to summarize the basic concepts of MR, its application and relevance in nephrology, and the methodological challenges and limitations as well as discuss the current guidelines for design and reporting. With reference to a clinically relevant example of examining the causal relationship between the estimated glomerular filtration rate and cancer, this review outlines the key steps to conducting an MR study, including the key considerations and potential pitfalls at each step. These include defining the clinical question, selecting the data sources, identifying and refining appropriate genetic variants by considering linkage disequilibrium and associations with potential confounders, harmonization of variants across data sets, validation of the genetic instrument by assessing its strength, estimation of the causal effects, confirming the validity of the findings, and interpreting and reporting results. Editor’s NoteResidual confounding is an ever-present impediment to inferring causality from observational data. In theory, Mendelian randomization effectively eliminates both measured and unmeasured confounders, making it a robust design for causal inference. Its proper execution, however, is not for the faint of heart. This review presents the foundation of Mendelian randomization while highlighting the methodological challenges and potential pitfalls at each step. It is a valuable entry point to conduct and critically appraise MR studies. Residual confounding is an ever-present impediment to inferring causality from observational data. In theory, Mendelian randomization effectively eliminates both measured and unmeasured confounders, making it a robust design for causal inference. Its proper execution, however, is not for the faint of heart. This review presents the foundation of Mendelian randomization while highlighting the methodological challenges and potential pitfalls at each step. It is a valuable entry point to conduct and critically appraise MR studies. Mendelian randomization (MR) studies are becoming increasingly popular in genetic epidemiology to assess the causal relationship between an exposure and an outcome of interest. In nephrology and transplantation, there have been an increasing number of MR studies focused on evaluating the causal relationships between risk factors and outcomes of kidney disease. For example, studies have assessed the causal relationships of novel biomarkers such as urinary uromodulin and hypertension with incident chronic kidney disease (CKD) or the effects of basic elements such as selenium on CKD risk.1Ponte B. Sadler M.C. Olinger E. et al.Mendelian randomization to assess causality between uromodulin, blood pressure and chronic kidney disease.Kidney Int. 2021; 100: 1282-1291Abstract Full Text Full Text PDF PubMed Scopus (15) Google Scholar,2Fu S. Zhang L. Ma F. et al.Effects of selenium on chronic kidney disease: a Mendelian randomization study.Nutrients. 2022; 14: 4458Crossref PubMed Scopus (5) Google Scholar Many of these relationships have been previously explored in observational studies with inconsistent or uncertain findings, owing to issues related to confounding biases, where unmeasured factors induce a spurious association between the exposure and the outcome. Previous methods such as inverse probability weighting of marginal structural models3Stephens-Shields A.A.-O. Spieker A.J. Anderson A. et al.Blood pressure and the risk of chronic kidney disease progression using multistate marginal structural models in the CRIC Study.Stat Med. 2018; 36: 4167-4181Crossref Scopus (7) Google Scholar and propensity score matching4Oh T.R. Choi H.S. Kim C.S. et al.Hyperuricemia has increased the risk of progression of chronic kidney disease: propensity score matching analysis from the KNOW-CKD study.Sci Rep. 2019; 9: 6681Crossref PubMed Scopus (76) Google Scholar have been used in observational studies to deal with confounding, but they are valid only under the assumption of no unmeasured confounding. In contrast, the randomized controlled trial (RCT) remains the criterion standard for the evaluation of causal relationships because it is an experimental design where participants are randomly assigned to the intervention (exposed) and control (unexposed) groups. This means that a difference in outcomes between these 2 groups should be attributable to the exposure of interest. However, an RCT is not feasible in all settings and for all exposures owing to ethical, resource, and logistical barriers. For example, an RCT cannot be conducted to investigate the impact of CKD on outcomes of interest because it would be unethical to assign a person to the harmful effects of CKD. Causal inference offers frameworks, theories, tools, and methods to examine the causal effects of exposures on outcomes.5Pearl J. An introduction to causal inference.Int J Biostat. 2010; 6: article 7Crossref PubMed Scopus (330) Google Scholar One such methodology is instrumental variable (IV) analysis, which uses IVs to estimate causal effects of exposures on outcomes without bias from measured and unmeasured confounders.6Greenland S. An introduction to instrumental variables for epidemiologists.Int J Epidemiol. 2000; 29: 722-729Crossref PubMed Scopus (685) Google Scholar MR methods are a type of IV analysis that uses genetic variants as IVs. These analyses estimate the effect of the genetic IV on both the exposure and the outcome to evaluate whether there is a causal effect of the exposure on the outcome. Figure 1 highlights the parallels between RCT and MR study designs. MR methods are valuable tools to investigate whether epidemiological observations reflect causal relationships. This review aims to summarize the basic concepts of MR, its application and relevance in nephrology, the methodological challenges and limitations, and the current guidelines for conducting and reporting. These are presented through the 6 key steps for undertaking an MR study, summarized in Figure 2. Table 17Cole S.R. Platt R.W. Schisterman E.F. et al.Illustrating bias due to conditioning on a collider.Int J Epidemiol. 2010; 39: 417-420Crossref PubMed Scopus (560) Google Scholar, 8VanderWeele T.J. Principles of confounder selection.Eur J Epidemiol. 2019; 34: 211-219Crossref PubMed Scopus (613) Google Scholar, 9Tennant P.A.-O.X. Murray E.J. Arnold K.A.-O. et al.Use of directed acyclic graphs (DAGs) to identify confounders in applied health research: review and recommendations.Int J Epidemiol. 2021; 50: 620-632Crossref PubMed Scopus (249) Google Scholar, 10Burgess S. Smith G.D. Davies N.M. et al.Guidelines for performing Mendelian randomization investigations (version 2; peer reviewed: 2 approved).Wellcome Open Res. 2020; 4 (article 186)Crossref Scopus (512) Google Scholar, 11Davies N.M. Holmes M.V. Davey Smith G. Reading Mendelian randomisation studies: a guide, glossary, and checklist for clinicians.BMJ. 2018; 362: k601Crossref PubMed Scopus (1263) Google Scholar, 12Sanderson E. Glymour M.M. Holmes M.V. et al.Mendelian randomization.Nat Rev Methods Primers. 2022; 2: 6Crossref PubMed Scopus (133) Google Scholar, 13Lawlor D.A. Harbord R.M. Sterne J.A.C. et al.Mendelian randomization: using genes as instruments for making causal inferences in epidemiology.Stat Med. 2008; 27: 1133-1163Crossref PubMed Scopus (2015) Google Scholar, 14Burgess S. Thompson S.G. CRP CHD Genetics CollaborationAvoiding bias from weak instruments in Mendelian randomization studies.Int J Epidemiol. 2011; 40: 755-764Crossref PubMed Scopus (825) Google Scholar provides a glossary of key terms and concepts used throughout this review. Table 215Bender R. Lange S. Adjusting for multiple testing—when and how?.J Clin Epidemiol. 2001; 54: 343-349Abstract Full Text Full Text PDF PubMed Scopus (2006) Google Scholar highlights some common pitfalls in MR studies.Table 1Glossary of key terms used in this primerTermDefinitionGeneral epidemiological concepts/terms Collider biasA collider is a variable that is a shared effect of 2 variables (e.g., exposure and outcome). Conditioning on a collider opens a collider path that transmits a spurious (noncausal) association, resulting in collider bias.7Cole S.R. Platt R.W. Schisterman E.F. et al.Illustrating bias due to conditioning on a collider.Int J Epidemiol. 2010; 39: 417-420Crossref PubMed Scopus (560) Google Scholar Confounding biasConfounding bias occurs when the analysis does not adjust for a common cause of the exposure and outcome of interest, inducing a spurious (noncausal) association between the exposure and the outcome.8VanderWeele T.J. Principles of confounder selection.Eur J Epidemiol. 2019; 34: 211-219Crossref PubMed Scopus (613) Google Scholar Directed acyclic graphA type of causal diagram that follows specific rules, which can be used to represent the knowledge, theories, and assumptions we have about the causal relationships between variables in a study.9Tennant P.A.-O.X. Murray E.J. Arnold K.A.-O. et al.Use of directed acyclic graphs (DAGs) to identify confounders in applied health research: review and recommendations.Int J Epidemiol. 2021; 50: 620-632Crossref PubMed Scopus (249) Google Scholar In a directed acyclic graph, arrows between variables (nodes) must be directed and single headed and represent the direction of causation between 2 variables. They must also be acyclic in that a variable cannot cause itself (i.e., there are no feedback loops from a variable back to itself).9Tennant P.A.-O.X. Murray E.J. Arnold K.A.-O. et al.Use of directed acyclic graphs (DAGs) to identify confounders in applied health research: review and recommendations.Int J Epidemiol. 2021; 50: 620-632Crossref PubMed Scopus (249) Google Scholar A causal path from the exposure to the outcome is one where all arrows flow in the same direction and represents a causal effect, which may be mediated by ≥1 explanatory variables on the pathway.9Tennant P.A.-O.X. Murray E.J. Arnold K.A.-O. et al.Use of directed acyclic graphs (DAGs) to identify confounders in applied health research: review and recommendations.Int J Epidemiol. 2021; 50: 620-632Crossref PubMed Scopus (249) Google Scholar Instrumental variableA variable that is associated with the exposure and is associated with the outcome only through its effect on the exposure (it is not associated with the outcome owing to confounding, and there is no causal path to the outcome except through the exposure).6Greenland S. An introduction to instrumental variables for epidemiologists.Int J Epidemiol. 2000; 29: 722-729Crossref PubMed Scopus (685) Google Scholar,10Burgess S. Smith G.D. Davies N.M. et al.Guidelines for performing Mendelian randomization investigations (version 2; peer reviewed: 2 approved).Wellcome Open Res. 2020; 4 (article 186)Crossref Scopus (512) Google Scholar Reverse causationThis occurs when the outcome of interest influences the exposure of interest (e.g., a disease or preclinical stage of disease affects the exposure).11Davies N.M. Holmes M.V. Davey Smith G. Reading Mendelian randomisation studies: a guide, glossary, and checklist for clinicians.BMJ. 2018; 362: k601Crossref PubMed Scopus (1263) Google ScholarConcepts and terms related to genetics and MR studies Allele (genetic) scoreAn approach that can be used in MR studies examining multiple genetic variants, whereby the variants are combined into a single score, which is used as a single genetic instrumental variable. The score can be unweighted (total number of risk-increasing alleles) or weighted (often using the estimated genetic effect size). The alternative is to treat the individual variants as separate instrumental variables and analyze them with meta-analysis.11Davies N.M. Holmes M.V. Davey Smith G. Reading Mendelian randomisation studies: a guide, glossary, and checklist for clinicians.BMJ. 2018; 362: k601Crossref PubMed Scopus (1263) Google Scholar Horizontal pleiotropyA type of pleiotropy where there is another causal pathway from the genetic variant to the outcome of interest aside from the pathway via the exposure of interest. This is a violation of the exclusion restriction assumption of MR and leads to bias.12Sanderson E. Glymour M.M. Holmes M.V. et al.Mendelian randomization.Nat Rev Methods Primers. 2022; 2: 6Crossref PubMed Scopus (133) Google Scholar Linkage disequilibriumThe correlation between certain genetic variants in the population, particularly those located close to each other within the same locus on the same chromosome. In MR studies, this may result in confounding if there is linkage disequilibrium between the genetic variant of interest and another genetic variant that is also associated with the outcome of interest.13Lawlor D.A. Harbord R.M. Sterne J.A.C. et al.Mendelian randomization: using genes as instruments for making causal inferences in epidemiology.Stat Med. 2008; 27: 1133-1163Crossref PubMed Scopus (2015) Google Scholar MRA research method that uses genetic variation to examine causal relationships between exposures and outcomes from observational data.10Burgess S. Smith G.D. Davies N.M. et al.Guidelines for performing Mendelian randomization investigations (version 2; peer reviewed: 2 approved).Wellcome Open Res. 2020; 4 (article 186)Crossref Scopus (512) Google Scholar One-sample MR studyA type of MR study where the variant-exposure and variant-outcome association estimates are derived from the same data source10Burgess S. Smith G.D. Davies N.M. et al.Guidelines for performing Mendelian randomization investigations (version 2; peer reviewed: 2 approved).Wellcome Open Res. 2020; 4 (article 186)Crossref Scopus (512) Google Scholar (see also Two-sample MR study). PleiotropyThe potential for genetic variants to have >1 phenotypic effect13Lawlor D.A. Harbord R.M. Sterne J.A.C. et al.Mendelian randomization: using genes as instruments for making causal inferences in epidemiology.Stat Med. 2008; 27: 1133-1163Crossref PubMed Scopus (2015) Google Scholar (see also Horizontal pleiotropy and Vertical pleiotropy). Population stratificationThis occurs when the sample used for MR includes different population groups that have different distributions of the genetic variant(s) of interest and different distributions of the phenotype/outcome of interest. This can introduce confounding into the relationship between the genetic variants and the outcome.13Lawlor D.A. Harbord R.M. Sterne J.A.C. et al.Mendelian randomization: using genes as instruments for making causal inferences in epidemiology.Stat Med. 2008; 27: 1133-1163Crossref PubMed Scopus (2015) Google Scholar SNPA genomic variation at a single base position in the DNA sequence.13Lawlor D.A. Harbord R.M. Sterne J.A.C. et al.Mendelian randomization: using genes as instruments for making causal inferences in epidemiology.Stat Med. 2008; 27: 1133-1163Crossref PubMed Scopus (2015) Google Scholar Two-sample MR studyA type of MR study where the variant-exposure and variant-outcome estimates are derived from separate data sources10Burgess S. Smith G.D. Davies N.M. et al.Guidelines for performing Mendelian randomization investigations (version 2; peer reviewed: 2 approved).Wellcome Open Res. 2020; 4 (article 186)Crossref Scopus (512) Google Scholar (see also Two-sample MR study). Vertical pleiotropyA type of pleiotropy where there is another phenotype on the causal pathway from genetic variant→exposure→outcome. This does not lead to bias in MR studies if the proximal phenotype of the exposure is correctly identified (see Step 3: selecting appropriate variants (from the variant-exposure data)).12Sanderson E. Glymour M.M. Holmes M.V. et al.Mendelian randomization.Nat Rev Methods Primers. 2022; 2: 6Crossref PubMed Scopus (133) Google Scholar Weak instrumentGenetic variants that are not strongly associated with the exposure.13Lawlor D.A. Harbord R.M. Sterne J.A.C. et al.Mendelian randomization: using genes as instruments for making causal inferences in epidemiology.Stat Med. 2008; 27: 1133-1163Crossref PubMed Scopus (2015) Google Scholar,14Burgess S. Thompson S.G. CRP CHD Genetics CollaborationAvoiding bias from weak instruments in Mendelian randomization studies.Int J Epidemiol. 2011; 40: 755-764Crossref PubMed Scopus (825) Google Scholar In a finite sample with a weak instrument, there is more likely to be chance variation in confounders that influence the phenotype between genotypic subgroups. As the strength of the instrument increases, phenotypic differences between the genotypic subgroups are more likely to be due to the genetic instrument than variations in confounders.14Burgess S. Thompson S.G. CRP CHD Genetics CollaborationAvoiding bias from weak instruments in Mendelian randomization studies.Int J Epidemiol. 2011; 40: 755-764Crossref PubMed Scopus (825) Google Scholar Weak instrument bias is therefore exacerbated by small sample sizes. In 1-sample MR studies, weak instruments lead to bias toward the observational association while in 2-sample MR studies they lead to bias toward the null.11Davies N.M. Holmes M.V. Davey Smith G. Reading Mendelian randomisation studies: a guide, glossary, and checklist for clinicians.BMJ. 2018; 362: k601Crossref PubMed Scopus (1263) Google Scholar,14Burgess S. Thompson S.G. CRP CHD Genetics CollaborationAvoiding bias from weak instruments in Mendelian randomization studies.Int J Epidemiol. 2011; 40: 755-764Crossref PubMed Scopus (825) Google ScholarMR, Mendelian randomization; SNP, single-nucleotide polymorphism. Open table in a new tab Table 2Common pitfalls of MR studies in nephrologyPitfallExplanation or examplesViolation of the “relevance” assumption•Selecting IVs from nonsignificant, suggestive, or nonreplicating GWAS loci•Selecting IVs associated with eGFR but not associated with renal function (e.g., variants affecting creatinine metabolism rather than clearance).Violation of the “exchangeability” assumption•Selecting IVs on the basis of GWAS with substantial or uncorrected genomic inflation.•Selecting IVs from regions of extended linkage disequilibrium (e.g., HLA region). Certain regions have been identified as common sites for linkage disequilibrium across distant variants.Horizontal pleiotropy•Selecting IVs with known effects on multiple organ systems, pathways, or traits (e.g., variants from the HLA or ABO loci)•Selecting IVs from loci that may have an independent causal effect on both the exposure and the outcome.Weak instruments•Using a single SNP or single locus IV (e.g., individual GWAS loci for eGFR or CKD have relatively weak effects and thus usually generate very weak instruments).•Not providing power analyses along with the negative MR findings (e.g., the use of weak instruments may lead to type II error).Overlapping data sets•GWAS for eGFR and CKD use a large number of existing population-based cohorts that have also been used for GWAS of cardiovascular and other traits (variant-exposure and variant-outcome estimates from overlapping data sets can lead to bias in 2-sample MR designs).Nonoverlapping populations•Selecting IVs from GWAS for an exposure performed in a specific ancestry or distinct population, and testing IVs’ association with the outcome in another nonoverlapping ancestral population.Lack of bidirectional analyses•Reverse causality may be missed if bidirectional hypotheses are not examined.Failure to account for multiple testing•Testing multiple IVs against multiple outcomes without penalty for multiple testing.•Multiple testing increases the risk of type I error.15Bender R. Lange S. Adjusting for multiple testing—when and how?.J Clin Epidemiol. 2001; 54: 343-349Abstract Full Text Full Text PDF PubMed Scopus (2006) Google Scholar•It is important to identify the primary analysis in an MR study. Multiple testing can also be adjusted for, lowering the significance threshold, by a number of techniques such as the Bonferroni correction method.CKD, chronic kidney disease; eGFR, estimated glomerular filtration rate; GWAS, genome-wide association study; HLA, human leukocyte antigen; IV, instrumental variable; MR, Mendelian randomization; SNP, single-nucleotide polymorphism. Open table in a new tab MR, Mendelian randomization; SNP, single-nucleotide polymorphism. CKD, chronic kidney disease; eGFR, estimated glomerular filtration rate; GWAS, genome-wide association study; HLA, human leukocyte antigen; IV, instrumental variable; MR, Mendelian randomization; SNP, single-nucleotide polymorphism. A key foundation of a well-conducted MR study using observational data is a well-defined research hypothesis about the causal effect of an exposure on the outcome of interest.16Hernán M.A. The C-word: scientific euphemisms do not improve causal inference from observational data.Am J Public Health. 2018; 108: 616-619Crossref PubMed Scopus (268) Google Scholar When planning an MR study, researchers must also consider whether they are solely aiming to test a causal hypothesis or whether they are additionally seeking to quantify the size of the effect of the exposure on the outcome.10Burgess S. Smith G.D. Davies N.M. et al.Guidelines for performing Mendelian randomization investigations (version 2; peer reviewed: 2 approved).Wellcome Open Res. 2020; 4 (article 186)Crossref Scopus (512) Google Scholar Where there is an interest in quantifying the effect, there are additional considerations and assumptions, as detailed below.10Burgess S. Smith G.D. Davies N.M. et al.Guidelines for performing Mendelian randomization investigations (version 2; peer reviewed: 2 approved).Wellcome Open Res. 2020; 4 (article 186)Crossref Scopus (512) Google Scholar Furthermore, as MR estimates reflect the effect of an exposure over a lifetime, the size of the estimate should be interpreted cautiously as it is unlikely to reflect the effect of modifying the exposure at a given time point. Causal questions can be represented using directed acyclic graphs (DAGs), which are a commonly used tool in causal inference.9Tennant P.A.-O.X. Murray E.J. Arnold K.A.-O. et al.Use of directed acyclic graphs (DAGs) to identify confounders in applied health research: review and recommendations.Int J Epidemiol. 2021; 50: 620-632Crossref PubMed Scopus (249) Google Scholar DAGs are diagrams that represent our knowledge, theories, hypotheses, and assumptions about the causal relationships between variables in the context of a research question.9Tennant P.A.-O.X. Murray E.J. Arnold K.A.-O. et al.Use of directed acyclic graphs (DAGs) to identify confounders in applied health research: review and recommendations.Int J Epidemiol. 2021; 50: 620-632Crossref PubMed Scopus (249) Google Scholar,17Greenland S. Pearl J. Robins J.M. Causal diagrams for epidemiologic research.Epidemiology. 1999; 10: 37-48Crossref PubMed Scopus (2775) Google Scholar Figure 3 shows a DAG for our example research question: “Is there a causal relationship between kidney function, measured by estimated glomerular filtration rate (eGFRcr), and cancer incidence?” with eGFRcr being our exposure of interest and cancer incidence being our outcome of interest. Although there may be indisputable evidence showing the increased risk of cancer incidence in kidney transplant recipients resulting from long-term immunosuppression use,18Stewart J.H. Vajdic C.M. van Leeuwen M.T. et al.The pattern of excess cancer in dialysis and transplantation.Nephrol Dial Transplant. 2009; 10: 3225-3231Crossref Scopus (153) Google Scholar, 19Vajdic C.M. McDonald S.P. McCredie M.R.E. et al.Cancer incidence before and after kidney transplantation.JAMA. 2006; 296: 2823-2831Crossref PubMed Scopus (908) Google Scholar, 20Kasiske B.L. Snyder J.J. Gilbertson D.T. Wang C. Cancer after kidney transplantation in the United States.Am J Transplant. 2004; 4: 905-913Abstract Full Text Full Text PDF PubMed Scopus (894) Google Scholar, 21Au E.H. Chapman J.R. Craig J.C. et al.Overall and site-specific cancer mortality in patients on dialysis and after kidney transplant.J Am Soc Nephrol. 2019; 30: 471-480Crossref PubMed Scopus (72) Google Scholar the effects of early- to moderate-stage CKD on cancer risk are less clear and existing findings are conflicting. Prior research has shown that the excess risk of cancer, particularly for urological and lung cancers, begins with eGFRcr < 60 ml/min per 1.73 m2 and with elevated albumin-creatinine ratio,22Wong G. Hayen A. Chapman J.R. et al.Association of CKD and cancer risk in older people.J Am Soc Nephrol. 2009; 20: 1341-1350Crossref PubMed Scopus (225) Google Scholar while others found that the excess risk of cancer was associated with early-stage CKD only if eGFR was estimated using cystatin C.23Lees J.S. Ho F. Parra-Soto S. et al.Kidney function and cancer risk: an analysis using creatinine and cystatin C in a cohort study.eClinicalMedicine. 2021; 38101030Abstract Full Text Full Text PDF PubMed Scopus (20) Google Scholar An individual patient data meta-analysis of observational studies and trials found no association between reduced eGFRcr and overall cancer risk,24Wong G. Staplin N. Emberson J. et al.Chronic kidney disease and the risk of cancer: an individual patient data meta-analysis of 32,057 participants from six prospective studies.BMC Cancer. 2016; 16: 488Crossref PubMed Scopus (69) Google Scholar but an increased risk of urinary tract, endocrine, and digestive tract cancers was observed in patients on dialysis.24Wong G. Staplin N. Emberson J. et al.Chronic kidney disease and the risk of cancer: an individual patient data meta-analysis of 32,057 participants from six prospective studies.BMC Cancer. 2016; 16: 488Crossref PubMed Scopus (69) Google Scholar In our example, as shown in Figure 3, there are many potential confounders that could cause reduced eGFRcr and cancer, such as socioeconomic position, health behaviors, and environmental exposures. Often, key confounders are not available from existing data sources or the measures we have may be limited in scope or validity, leaving observational analyses vulnerable to residual confounding. An IV is a variable that is associated with the exposure and is associated with the outcome only through its effect on the exposure (it is not associated with the outcome owing to confounding, and there is no causal path to the outcome except through the exposure).6Greenland S. An introduction to instrumental variables for epidemiologists.Int J Epidemiol. 2000; 29: 722-729Crossref PubMed Scopus (685) Google Scholar That is, the only path on a DAG from the IV to the outcome is a causal path via the exposure.6Greenland S. An introduction to instrumental variables for epidemiologists.Int J Epidemiol. 2000; 29: 722-729Crossref PubMed Scopus (685) Google Scholar,10Burgess S. Smith G.D. Davies N.M. et al.Guidelines for performing Mendelian randomization investigations (version 2; peer reviewed: 2 approved).Wellcome Open Res. 2020; 4 (article 186)Crossref Scopus (512) Google Scholar Provided that the IV assumptions below are satisfied, we can estimate the causal effect of the exposure on the outcome without bias from measured and unmeasured confounders of the exposure-outcome relationship by using the association of the IV with both the exposure and the outcome.6Greenland S. An introduction to instrumental variables for epidemiologists.Int J Epidemiol. 2000; 29: 722-729Crossref PubMed Scopus (685) Google Scholar In MR, we use ≥1 genetic variants that are robustly associated with a phenotypic exposure as an IV. Although different types of genetic variants can be used as IVs in MR, single-nucleotide polymorphisms (SNPs) are often chosen,12Sanderson E. Glymour M.M. Holmes M.V. et al.Mendelian randomization.Nat Rev Methods Primers. 2022; 2: 6Crossref PubMed Scopus (133) Google Scholar as in our example where we focus on SNPs associated with eGFR. Genetic variants can be used as IVs to overcome exposure-outcome confounding because of the ran