Abstract

Background: Mendelian randomization (MR) studies investigate the effect of genetic variation in levels of an exposure on an outcome, thereby using genetic variation as an instrumental variable (IV). We provide a meta-epidemiological overview of the methodological approaches used in MR studies, and evaluate the discussion of MR assumptions and reporting of statistical methods.

Methods: We searched PubMed, Medline, Embase and Web of Science for MR studies up to December 2013. We assessed (i) the MR approach used; (ii) whether the plausibility of MR assumptions was discussed; and (iii) whether the statistical methods used were reported adequately.

Results: Of 99 studies using data from one study population, 32 used genetic information as a proxy for the exposure without further estimation, 44 performed a formal IV analysis, 7 compared the observed with the expected genotype-outcome association, and 1 used both the latter two approaches. The 80 studies using data from multiple study populations used many different approaches to combine the data; 52 of these studies used some form of IV analysis; 44% of studies discussed the plausibility of all three MR assumptions in their study. Statistical methods used for IV analysis were insufficiently described in 14% of studies.

Conclusions: Most MR studies either use the genotype as a proxy for exposure without further estimation or perform an IV analysis. The discussion of underlying assumptions and reporting of statistical methods for IV analysis are frequently insufficient. Studies using data from multiple study populations are further complicated by the combination of data or estimates. We provide a checklist for the reporting of MR studies.

Key Messages

  • The specific methods used in Mendelian randomization studies vary widely.

  • These methods broadly fall into three categories: (i) using genetic information as a proxy for the exposure without further estimation; (ii) performing an instrumental variable analysis; (iii) comparing the observed with the expected genotype-outcome association.

  • Mendelian randomization studies frequently insufficiently discuss underlying assumptions or report statistical methods for IV analysis.

  • A checklist for the reporting of Mendelian randomization studies is provided.

Introduction

Observational studies are limited in their ability to identify whether exposures are causally related to disease occurrence or other outcomes. Adjustment for confounding is only possible for those factors which are identified and measured and will inevitably be incomplete: some degree of residual confounding will always remain. Reverse causation, an effect of the outcome on the studied exposure, may also explain associations found in an observational study.1,2 An approach which can circumvent both reverse causation (as first proposed in 1986)3 and residual confounding in order to establish the causal effect of the exposure on the outcome is to investigate the effect of genetic variation in levels of the exposure on the outcome. This approach has come to be known as Mendelian randomization over the past decade.2 The random allocation of genetic variants from parents to offspring means these variants will generally be unrelated to other factors which affect the outcome.1 Furthermore, associations between the genotype and the outcome will not be affected by reverse causation because disease does not affect genotype.1

Mendelian randomization studies use genetic variation as an instrumental variable (IV) and must fulfil instrumental variable assumptions. Applied to Mendelian randomization, these assumptions are that (i) the genotype is associated with the exposure; (ii) the genotype is associated with the outcome through the studied exposure only (exclusion restriction assumption); and (iii) the genotype is independent of other factors which affect the outcome (independence assumption).4 Potential threats to the validity of these assumptions, such as population stratification, linkage disequilibrium and pleiotropic effects, are discussed in detail elsewhere.1,5

These general principles of Mendelian randomization are increasingly used in aetiological research, but the specific methods used in these studies can vary widely. In this study we review the methodology used in studies from the past 10 years which were described by the authors as Mendelian randomization studies. We provide an overview of the use of the different approaches to Mendelian randomization and, where applicable, the specific statistical methods used for estimation. We evaluate whether the plausibility of the Mendelian randomization assumptions is discussed. Further, we evaluate whether the statistical methods used are sufficiently described (including how the confidence interval was obtained) for those studies which perform an instrumental variable analysis or compare the observed and expected genotype-outcome association.

Methods

Search strategy and inclusion criteria

We searched PubMed, Medline, Embase and Web of Science for studies containing the term ‘Mendelian randomization’ or ‘genetic instrumental variable’ or a related term (e.g. ‘genetic instrument’)from 1 January 2003 to 31 December 2013. The full search strategies for each of the databases are included in Supplementary Methods, available as Supplementary data at IJE online. We excluded publications that; (i) were conference abstracts, letters, commentaries, editorials, reviews, study proposals or theoretical papers; (ii) did not use Mendelian randomization (i.e. did not state Mendelian randomization or a genetic instrumental variable was used in the text, abstract or title and did not include ‘Mendelian randomization’ or ‘genetic instrumental variable’ or a related term as a keyword); (iii) identified potential genetic instruments for future Mendelian randomization studies; (iv) were primarily methodological, using an application of Mendelian randomization as an example; or (v) were published in a health economics journal (rather than a biomedical journal).

Classification of Mendelian randomization approach used

First we classified publications into studies which used data from a single study population and studies which used data from multiple study populations. We then classified included studies according to their general Mendelian randomization approach: i.e. how they utilized the genetically determined variation in exposure.

  • For studies performed in a single study population we identified the following three main approaches:

    • Use of genetic variation as a proxy for the exposure, without further estimation.

      •    These studies investigate the association between a genotype (which affects the exposure) and the outcome. No comparison is made with the expected association between this genotype and the outcome, and no IV estimate of the effect of the exposure on the outcome is obtained.

    • Comparison of the observed and expected genotype-outcome associations.

      •    These studies compare the observed association between the genotype and the outcome with the association which would be expected if the observed exposure-outcome association were causal. This expected association is calculated by multiplying the observed genotype-exposure association with the observed exposure-outcome association (sometimes termed the ‘triangulation’ approach, although this is not a specific term); see Figure 1 for an illustration and further explanation. The confidence interval of the expected genotype-outcome association can be estimated analytically or using bootstrap techniques.

    • Formal instrumental variable analysis using genetic variation as the instrument.

      •    These studies perform a formal IV analysis to obtain a causal estimate of the effect of genetically determined variation in the exposure on the outcome. Different statistical techniques can be used for this purpose, as we will further explain below.

  • We classified the studies including more than one study population into the following pre-specified main categories:

    • Pooling of the data, followed by any of the approaches Ai‐iii listed above.

    • IV analysis in each of the study populations, followed by a meta-analysis.

    • Meta-analysis using the genotype as a proxy for the exposure, without further estimation.

    • Meta-analyses of the genotype-exposure, exposure-outcome and genotype-outcome associations, followed by comparison of observed and expected genotype-outcome associations (as in approach Aii).

    • Meta-analyses of the genotype-exposure and genotype-outcome associations, followed by a Wald-type/ratio estimate (see Didelez et al. for a description of Wald-type estimators).6

    • Data analysed separately for more than one population, followed by any of the approaches Ai‐iii.

Figure 1.

Diagram of the approach used by Mendelian randomization studies which compare the observed genotype-outcome association with the expected genotype-outcome association. β1, regression coefficient of the genetic variant-exposure association. β2, regression coefficient of the exposure-outcome association. β3OBS, observed regression coefficient of the genetic variant-outcome association. β3EXP, expected regression coefficient of the genetic variant-outcome association. The point estimate of β3EXP is calculated as follows: β3EXP = β1 · β2. The confidence interval of the expected genotype-outcome association can be estimated analytically or using bootstrap techniques.

Figure 1.

Diagram of the approach used by Mendelian randomization studies which compare the observed genotype-outcome association with the expected genotype-outcome association. β1, regression coefficient of the genetic variant-exposure association. β2, regression coefficient of the exposure-outcome association. β3OBS, observed regression coefficient of the genetic variant-outcome association. β3EXP, expected regression coefficient of the genetic variant-outcome association. The point estimate of β3EXP is calculated as follows: β3EXP = β1 · β2. The confidence interval of the expected genotype-outcome association can be estimated analytically or using bootstrap techniques.

Further categories were added for those studies which did not fall into any of the above categories.

Assessment of discussion of Mendelian randomization assumptions

Regardless of the approach used, Mendelian randomization studies rely on three main assumptions, as briefly mentioned in the introduction.

  • The genotype is associated with the exposure.

    •   This assumption can and should be verified in the data. Reporting guidelines for IV analyses recommend the use of the partial F-statistic as a measure of the strength of the association between the IV and the exposure.7,8 It encompasses information on the strength of the instrument and on the number of observations in the analysis.9 We assessed whether studies reported the strength of the genotype-exposure association in the data using a partial F-statistic or using another measure (e.g. mean difference in exposure by genotype). If not, we assessed whether they reported the strength of this association from literature.

  • The genotype is associated with the outcome through the studied exposure only (exclusion restriction assumption).

    •   This assumption is violated if the genotype has multiple (pleiotropic) effects, if a nearby variant with which it is in linkage disequilibrium affects the outcome in other ways than through the exposure of interest, or if developmental canalization occurs.1 For all studies we evaluated whether the plausibility of this assumption was discussed. Mentioning the assumption in general terms was not deemed sufficient: a specific discussion of its plausibility in the particular study was required.

  • The genotype is independent of other factors which affect the outcome (independence assumption).

    •   This assumption is violated if subgroups in the study population have both different genotype frequencies and different distributions of the outcome (population stratification).1 It is also violated if there is an association between the genotype used as an instrument and confounders. For all studies we assessed whether the association between the genetic instrument and measured confounders was reported, as recommended in IV reporting guidelines.7 Furthermore, we assessed whether potential associations of the genotype with unmeasured confounders were discussed and/or population stratification was discussed. Again, a specific discussion of the plausibility of the assumption in the particular study was required.

Assessment of reporting of statistical analysis

This section only applies to the studies which used the IV approach or the observed-expected approach, because using genetic variation as a proxy for the exposure without further estimation does not involve any special statistical methods.

  • For studies which obtained an IV estimate of the effect of the exposure on the outcome, we determined which statistical method was used and assessed whether it was described sufficiently and whether a confidence interval was reported. A frequently used IV method is two-stage least squares analysis. This involves two stages of linear regression. The first stage is a linear regression with the exposure as the dependent variable and the instrument (genotype) as the independent variable, which is then used to obtain predicted exposure levels based on the instrument. The second stage is a regression with the outcome as the dependent variable and these genetically predicted exposure levels as the independent variable. Software for two-stage least squares regression takes into account the errors in both stages of the analysis to give a correct confidence interval. Additionally, we determined the type of outcome investigated (continuous, binary, time-to-event) and for binary outcomes what kind of target parameter was estimated (risk difference, odds ratio, relative risk, probit coefficient). We also determined whether a statistical test was used to compare the IV estimate with the ‘conventional’ estimate of the effect of the exposure on the outcome, what type of genetic instrument was used [one single nucleotide polymorphism (SNP) or allele, multiple SNPs in separate analyses, multiple SNPs in a single analysis, combinations of SNPs e.g. haplotypes or a genetic risk score] and for those studies which used multiple SNPs in a single analysis, whether weak instrument bias was discussed. In the IV studies within one study population, we also determined whether the genetic variant used as an instrument was identified or selected in the same population or if the weights for a weighted genetic risk score were derived in the same population.

  • For studies comparing the observed and expected genotype-outcome association, we assessed whether the method used to obtain a point estimate of the expected genotype-outcome association was described. If the description was such that calculation of this point estimate should be possible using the data provided, we assessed whether the point estimate corresponded to our calculations (only in those studies within one population). Further, we assessed whether a confidence interval for the expected genotype-outcome association was reported, whether the method used to obtain this confidence interval was described, and whether the confidence interval incorporated the variance of both the genotype-exposure association and the exposure-outcome association.

Results

Our search returned 1911 hits, of which 594 hits remained after exclusion of conference abstracts and duplications. After reviewing the title and abstract and if necessary the full-text article, a further 415 records were excluded for reasons listed in the flowchart in Figure 2, resulting in 179 eligible Mendelian randomization studies. Of these 179 studies, 99 studies used data from a single study population for their main analyses,10–108 and 80 studies used data from more than one study population (Table 1).109–188 The included studies were published between May 2005 and December 2013. An overview of the exposures studied and the genetic instruments used is presented in Supplementary Table 1 (available as Supplementary data at IJE online). The most frequently studied exposures were C-reactive protein (29 studies) and adiposity measures such as body mass index, fat mass and percentage body fat (25 studies).

Figure 2.

Summary of literature search.

Figure 2.

Summary of literature search.

Table 1.

Approaches used in Mendelian randomization studies

Data from 1 study population (n = 99) Refs 
1. Genotype used as a proxy for exposure, without further estimationa 38 10–47 
2. Comparison of observed and expected genotype-outcome association 10 48–57 
3. IV analysisb 48 58–105 
4. Comparison of observed and expected genotype-outcome association and IV analysis 106 
5. Unclear 107,108 
Data from more than 1 study population (n = 80)  
 
1. Data pooled, then analysed   
 a. Genotype used as a proxy for exposure, without further estimation 109,110 
 b. Comparison of observed and expected genotype-outcome associationa 111–113 
 c. IV analysisb 14 114–127 
 d. Comparison of observed and expected genotype-outcome association and IV analysis 128–134 
2. IV analyses, then meta-analysis 10 135–144 
3. Meta-analysis using genotype as a proxy for exposure, without further estimation 15 134–136,145–157 
4. Meta-analyses* followed by comparison of observed and expected genotype-outcome associationc 13 139,144,157–167 
5. Meta-analyses* followed by a Wald-type/ratio estimate 168–176 
6. Data analysed and reported separately for more than 1 population   
 a. Genotype used as a proxy for exposure, without further estimationc 177,179,180 
 b. IV analysis 178,181,182 
7. Multivariate meta-analysis 155,183 
8. Bayesian meta-analysis 184 
9. Separate study IV-analysis 185 
10. Meta-analysis of gene-exposure association, then ratio estimate, then meta-analysis 186 
11. Other/unclear** 187,188 
Data from 1 study population (n = 99) Refs 
1. Genotype used as a proxy for exposure, without further estimationa 38 10–47 
2. Comparison of observed and expected genotype-outcome association 10 48–57 
3. IV analysisb 48 58–105 
4. Comparison of observed and expected genotype-outcome association and IV analysis 106 
5. Unclear 107,108 
Data from more than 1 study population (n = 80)  
 
1. Data pooled, then analysed   
 a. Genotype used as a proxy for exposure, without further estimation 109,110 
 b. Comparison of observed and expected genotype-outcome associationa 111–113 
 c. IV analysisb 14 114–127 
 d. Comparison of observed and expected genotype-outcome association and IV analysis 128–134 
2. IV analyses, then meta-analysis 10 135–144 
3. Meta-analysis using genotype as a proxy for exposure, without further estimation 15 134–136,145–157 
4. Meta-analyses* followed by comparison of observed and expected genotype-outcome associationc 13 139,144,157–167 
5. Meta-analyses* followed by a Wald-type/ratio estimate 168–176 
6. Data analysed and reported separately for more than 1 population   
 a. Genotype used as a proxy for exposure, without further estimationc 177,179,180 
 b. IV analysis 178,181,182 
7. Multivariate meta-analysis 155,183 
8. Bayesian meta-analysis 184 
9. Separate study IV-analysis 185 
10. Meta-analysis of gene-exposure association, then ratio estimate, then meta-analysis 186 
11. Other/unclear** 187,188 

Some studies used multiple approaches in non-identical sets of study populations.

aTwo studies also performed an IV analysis for which it was unclear how the data were combined.112,113

bOne study performed some of the analyses in a single study population.118

cTwo studies also performed an IV analysis in a single study population.158,177

*Meta-analyses of genotype-exposure, exposure-outcome and/or genotype-outcome associations.

**One study first investigated the genotype-outcome association and then performed further analyses for which the approach was unclear.187 One study used a ‘likelihood-based method for combining summarised genetic association estimates’.188

Of the 99 studies which used data from a single study population, 38 studies (38%) used the genetic information as a proxy for the exposure by investigating the genotype-outcome association without further estimation of either the causal effect of the exposure on the outcome or of the expected genotype-outcome association (Table 1); 48 studies (48%) used IV analysis to estimate the effect of genetically determined variation in exposure levels on the outcome. Ten studies compared the observed association between the genotype and the outcome with the expected association based on the genotype-exposure association and the exposure-outcome association. One study used both these latter two approaches. For two studies we could not categorize the methods used into any of the aforementioned approaches.

Of the 80 studies which used data from multiple study populations, 26 (33%) studies pooled the data from the different studies and subsequently analysed the pooled data (Table 1). Ten studies performed an IV analysis in the different studies followed by a meta-analysis; 41 studies (51%) first performed a meta-analysis of one or more of the genotype-exposure, exposure-outcome and genotype-outcome associations, 26 of which subsequently used these meta-analysed associations for further estimation of either the causal effect of the exposure on the outcome or of the expected genotype-outcome association. In total, 52 studies (65%) used some form of IV analysis to obtain a causal effect of the exposure on the outcome. A further 23 studies compared the observed and expected genotype-outcome associations.

Table 2 summarizes the reporting of the Mendelian randomization assumptions. Reporting of assumptions was assessed in 178 studies, because the design of one study was so different from the general Mendelian randomization design that the assumptions could not be assessed. A total of 37 out of 98 studies (38%) which used a single study population and 42 out of 80 studies (53%) which used multiple study populations explicitly discussed the plausibility of all three Mendelian randomization assumptions in the context of their study.

Table 2.

Reporting of Mendelian randomization assumptions

Criteria
 
1 Study population (n = 98)* >1 Study population (n = 80) 
Strength of genetic instrument-exposure association (assumption 1)   
 Verified in data using F-statistic 33 26 
 Otherwise verified in data (e.g. using risk difference or odds ratio) 53 45 
 Reported from literature 
 Not reported 5 
Plausibility of exclusion restriction assumption discussed (assumption 2) 56 55 
Independence assumption (assumption 3)   
 Instrument-confounder associations shown & assumption further discussed theoreticallya 20 16 
 Instrument-confounder associations shown, assumption not further discussed 30 21 
 Investigation of instrument-confounder associations mentioned, not shown & assumption further  discussed theoretically 
 Investigation of instrument-confounder associations mentioned, not shown & assumption not further  discussed 
 Plausibility of assumption theoretically discussed only 18 
 Plausibility of assumption not discussed 29 25 
Criteria
 
1 Study population (n = 98)* >1 Study population (n = 80) 
Strength of genetic instrument-exposure association (assumption 1)   
 Verified in data using F-statistic 33 26 
 Otherwise verified in data (e.g. using risk difference or odds ratio) 53 45 
 Reported from literature 
 Not reported 5 
Plausibility of exclusion restriction assumption discussed (assumption 2) 56 55 
Independence assumption (assumption 3)   
 Instrument-confounder associations shown & assumption further discussed theoreticallya 20 16 
 Instrument-confounder associations shown, assumption not further discussed 30 21 
 Investigation of instrument-confounder associations mentioned, not shown & assumption further  discussed theoretically 
 Investigation of instrument-confounder associations mentioned, not shown & assumption not further  discussed 
 Plausibility of assumption theoretically discussed only 18 
 Plausibility of assumption not discussed 29 25 

*Reporting of assumptions was not assessed in one study, because its design was vastly different from the general design of a Mendelian randomization study. The total number of studies within 1 study population is therefore 98.

aPotential association with unmeasured confounders discussed and/or population stratification discussed.

Two studies reported a P-value only.

Among the studies which performed an IV analysis, those using a single study population most frequently studied a continuous outcome, whereas those using multiple study populations most frequently studied a binary outcome and estimated an odds ratio (Table 3). The statistical methods used in these formal IV studies are shown in Table 4. Two-stage least squares (2‐SLS) regression was the most common method used in studies within one study population (n = 26, 53%). Ten studies within multiple study populations also used this method. One study used 2‐SLS with a binary outcome, but it did not mention whether heteroskedasticity robust standard errors were used.69 Among the studies which used multiple study populations, a Wald-type or ratio estimator was most frequently used (n = 16, 31%). The method used to obtain the confidence interval for the ratio estimate was a Taylor series expansion (termed the delta method138,141,168,169,186 or Taylor expansion170), Fieller method,120,176–178 or was not described. Three studies in a single study population also used a Wald-type/ratio estimator, but two of these studies did not report a confidence interval. Other methods used were control functions (n = 8 in total), IV probit regression (n = 4), generalized method of moments (n = 8), generalized least squares regression (n = 5), quasi-likelihood and variance function (n = 4) and a two-stage approach with a linear first stage and a logistic second stage (n = 5). Four of the studies which used this last approach did not report how the correct confidence interval was obtained,125,133,135,158 and the fifth used a sandwich estimator.114 The IV method was insufficiently described in 14 studies. In six of these studies there was a discrepancy between the statistical method reportedly used (2‐SLS) and the effect estimate reported (OR).100,128,129,130,131,134 Another study seemingly did not take into account the variance of the genotype-exposure association in the variance of the IV estimate, which would result in too narrow a confidence interval.101

Table 3.

Types of outcome and parameters estimated in IV Mendelian randomization studies

Type of outcome 1 Study population (n = 49) Refs >1 Study population (n = 52) Refs 
Continuous 37 58–68,70–89,94,98,103–106 14 114–116,118,126,135–139,168,170,182,185 
Binary     
 Risk difference 69,93,99 – 
 Relative risk 81,102 118,124 
 Odds ratio 67,82,88,90–92,100 37 112–114,117,119–121,123,125,128–135,139–144,155,158,168,169,171–176,178,181,183,184,186 
 Probit coefficient 83 122 
Time-to-event 95–97,101 124,138,168,177,178 
Unclear – 127 
Type of outcome 1 Study population (n = 49) Refs >1 Study population (n = 52) Refs 
Continuous 37 58–68,70–89,94,98,103–106 14 114–116,118,126,135–139,168,170,182,185 
Binary     
 Risk difference 69,93,99 – 
 Relative risk 81,102 118,124 
 Odds ratio 67,82,88,90–92,100 37 112–114,117,119–121,123,125,128–135,139–144,155,158,168,169,171–176,178,181,183,184,186 
 Probit coefficient 83 122 
Time-to-event 95–97,101 124,138,168,177,178 
Unclear – 127 

The total number of types of outcome and parameters estimated exceeds the total number of studies because some studies included multiple types of outcomes.

Table 4.

Statistical methods used in the instrumental variable studies

Method 1 Study population (n = 49) Refs >1 Study population* (n = 52) Refs 
Two-stage least squares 26 58–83 10 114–116,118,135–139,182 
Instrumental variable regression in Stata, not further specified (2‐SLS, GMM or LIML) 84–88 – 
Control function 81,82,89–92 139,143 
Instrumental variable probit regression 67,83,93 122 
GMM 94,98 – 
Multiplicative GMM – 117–121,124 
Generalized least squares regression 95 112,113,123,132 
Two-stage: linear first stage, logistic second stage – 114,125,133,135,158 
Quasi-likelihood and variance function 88 140,144,181 
Ratio/Wald-type estimator 99 17 120,124,138,141,142,168–178,186 
Ratio/Wald-type estimator without confidence interval 96,97 – 
Insufficiently described/unclear 100–106 126–131,134 
Otherb – 155,183–185 
Method 1 Study population (n = 49) Refs >1 Study population* (n = 52) Refs 
Two-stage least squares 26 58–83 10 114–116,118,135–139,182 
Instrumental variable regression in Stata, not further specified (2‐SLS, GMM or LIML) 84–88 – 
Control function 81,82,89–92 139,143 
Instrumental variable probit regression 67,83,93 122 
GMM 94,98 – 
Multiplicative GMM – 117–121,124 
Generalized least squares regression 95 112,113,123,132 
Two-stage: linear first stage, logistic second stage – 114,125,133,135,158 
Quasi-likelihood and variance function 88 140,144,181 
Ratio/Wald-type estimator 99 17 120,124,138,141,142,168–178,186 
Ratio/Wald-type estimator without confidence interval 96,97 – 
Insufficiently described/unclear 100–106 126–131,134 
Otherb – 155,183–185 

The total number of statistical methods exceeds the total number of studies because some studies investigated multiple statistical methods.

2‐SLS, two-stage least squares; GMM, generalized method of moments; LIML, limited-information maximum likelihood.

*Including the two studies which used multiple study populations, but performed the IV analysis in a single study population.

bSee Table 1.

Of the101 studies which used one of the approaches which yields an IV estimate, 48 reported tests of the difference between the IV estimate and the conventional estimate: the most commonly used were (a variant of) the Durbin-Wu-Hausman test (29 studies),58–60,62,64–66,68,70–72,74,75–77,79–87,94,98,105,116,139 and (a variant of) the Bland-Altman test(10 studies).112,113,117,119,128–132,134 The types of genetic instrument used (e.g. a single SNP or a genetic risk score) in the IV analysis studies are listed in Supplementary Table 2 (available as Supplementary data at IJE online). Of the 25 studies which used multiple SNPs in a single analysis, 13 mentioned weak instrument bias, with two studies very specifically discussing it in relation to using multiple instruments.117,185 Of the 49 studies which used IV methods and were performed in one study population, 14 evidently identified or selected the genetic variant used as an instrument in the same population or derived weights for a weighted genetic risk score in the same population.61,62,65,66,70,71,82,85,87,91,93,102,104,105

In 3 of the 11 studies comparing the observed gene-outcome association with the expected gene-outcome association in one study population we could not reconstruct the point estimate of the expected association from the data.48,50,57 Four studies did not report a confidence interval for the expected genotype-outcome association.52,54,56,106 In a further five studies, the methods used to calculate this confidence interval were unclear,48–50,53,57 and in one study only the error in the exposure-outcome association seemed to have been taken into account in the calculation of this confidence interval.55 Only one study adequately described the methods used to obtain the point estimate and confidence interval (bootstrapping) of the expected genotype-outcome association.51 In the 23 studies which employed this approach using more than one study population, three only took into account the error in the exposure-outcome association and not the error in the genotype-exposure association,159,164,165 and 16 studies did not describe how the confidence interval was obtained.

Discussion

Most Mendelian randomization studies either performed some form of IV analysis (49% of studies within one study population and 65% of studies within multiple study populations) or used the genotype as a proxy for the exposure without further estimation. A third approach used less frequently was to compare the observed genotype-outcome association with the expected genotype-outcome association. Although validity of the three main Mendelian randomization assumptions is required regardless of the approach used, only 44% of studies adequately discussed the plausibility of these assumptions. The methods used to obtain an IV estimate were not always adequately described. For those studies which were performed using multiple study populations, the range of approaches used was very broad, because of further differentiation according to the way the data from the different studies were combined. Here we will discuss our findings and propose recommendations for the reporting of Mendelian randomization studies.

To our knowledge there is one paper which previously reviewed MR studies, which included a much smaller number of studies. Its main focus was on whether the Mendelian randomization studies reported results that were compatible with a causal association, which was the case for over half of their reviewed studies.189 In contrast, our review focused on the approach used and on the discussion of the assumptions and the reporting of the statistical methods used. The previous review also noted that many studies applied IV analysis to a binary outcome, using methods which had not quite been validated,189 which is an issue which we will also discuss later.

Our meta-epidemiological study has several limitations. With respect to study selection, we investigated what methods were used in studies stating that they used Mendelian randomization or that they used a genetic IV. Importantly, we were unable to include studies which apply the same principles without using the term Mendelian randomization or genetic IV because these could not feasibly be found using a systematic search strategy. We do not know to what extent our results apply to these studies, but suspect the discussion of Mendelian randomization assumptions in particular is likely to be insufficient in many of these studies. Importantly, the focus of our review was on the quality of reporting of methods used in Mendelian randomization studies. We did not assess whether the statistical method used to obtain an IV estimate was actually appropriate. We investigated whether the statistical method used was adequately described, whether it was consistent with the estimates reported and if any evident mistakes were made. Similarly, we focused on whether plausibility of MR assumptions was discussed, not on whether we considered them likely to hold.

With regard to the Mendelian randomization approach used, we found that a majority of studies performed some form of IV analysis, but a substantial proportion of studies used the genotype as a proxy for the exposure without performing a formal IV analysis. This raises the question whether either of these approaches, or the third option of comparing the observed and expected genotype-outcome association, should be preferred. This depends on the aim of the study: for a test of causality, testing the presence of a genotype-outcome association is sufficient.1,190 Often the aim will be a quantification of the causal effect of the exposure on the outcome. We note that IV analysis is more suited to this aim than a comparison of the observed and the expected effect of the genotype on the outcome, although some may find the latter approach more intuitive. Showing the association between the genotype (or genetic score) and the outcome is always advisable as it increases the transparency of the study by showing the data as they are. Further analyses can subsequently be undertaken.191 When considering whether a formal IV analysis is appropriate, further aspects of the underlying biology of the genotype-phenotype association need to be taken into account to avoid misleading inferences.192 A recent paper discusses a number of situations in which a formal IV analysis may give biased results, but a Mendelian randomization approach looking only at the genotype-outcome association can validly be used as a test of causality.190 Another recent paper specifically discusses smoking as an example of an exposure for which the measurement does not fully capture the underlying exposure, which gives a biased estimate of the effect of the measured exposure on the outcome if an IV analysis is performed in a Mendelian randomization study.4

With regard to the discussion of the Mendelian randomization assumptions, we found that fewer than half of studies adequately discussed all three assumptions. Some studies did mention what the assumptions are and how they can be violated in general terms, but did not discuss how plausible the assumptions were for the specific setting of their study. An aspect of the assumptions which can be evaluated using the data is whether there is an association between the genetic instrument and measured confounders. This may be more difficult for studies which use multiple study populations, but an effort to obtain this information from those studies in which it is available is warranted. Among the studies which performed an IV analysis in a single study population, we identified 14 studies in which SNPs were detected or selected, or genetic risk score weights were derived in that same study population. This can bias Mendelian randomization estimates.116,193 The number of studies in which we found this to have occurred may be an underestimation, because some study populations are used for multiple Mendelian randomization studies and the later studies may not report the detection of SNPs in a previous study in the same population.

With regard to the IV methods used, we found that two-stage least squares regression and a Wald-type/ratio estimator were the most commonly used methods. We also found that a considerable number of the Mendelian randomization studies which used IV methods estimated an odds ratio or risk ratio, especially in those studies which used data from multiple study populations. However, which methods are appropriate for IV estimation of causal odds ratios or risk ratios is a methodological challenge of IV analysis that has not yet been fully resolved. Several methodological studies have investigated this issue in recent years.194,194–198 One of the reviewed MR studies mentioned that the Wald-type estimator used to estimate an odds ratio was an approximate method.169 The properties and limitations of these IV methods used to estimate a causal odds ratio deserve more attention in the Mendelian randomization studies in which they are used.

Overall, we conclude from our review the standard of reporting of Mendelian randomization studies should be improved. Existing guidelines and recommendations for the reporting of IV analyses largely apply to Mendelian randomization studies (the extent depending on the Mendelian randomization approach used).7,8 In addition to these recommendations, we have formulated a checklist of Mendelian randomization-specific reporting recommendations in Box 1.

Box 1. Proposed checklist for reporting Mendelian randomization studies

Methods

  • If an expected genotype-outcome association is calculated, report how this was calculated and how the confidence interval was obtained. Take into account the variance of both the genotype-exposure and the exposure-outcome association.

  • If an instrumental variable analysis is performed, report in detail which method was used and how the confidence interval was obtained. For non-standard instrumental variable methods (e.g. methods used to estimate an odds ratio), discuss the properties of these methods.

  • If data from multiple populations are used, clearly explain how and at what stage the data/estimates were combined.

Results

  • Report the strength of the association between the genetic instrument and the exposure, using a partial F-statistic if possible.

  • Show the association between the genetic instrument and measured confounders. If multiple study populations are used, show this for those populations for which this information is available.

  • Report the association of the genotype and the outcome.

Discussion

  • Discuss the plausibility of the second and third instrumental variable assumptions in the specific setting of the study: could pleiotropy, linkage disequilibrium, canalization, population stratification or unmeasured confounding of the genotype-outcome relation affect results in this study?

In conclusion, studies stating that they perform a Mendelian randomization study within one study population broadly fall into three categories: studies using a genotype as a proxy for exposure without further estimation; studies performing IV analysis using a genotype as an instrument; and studies comparing observed and expected genotype-outcome associations. Plausibility of underlying Mendelian randomization assumptions is not always discussed, but as these assumptions are crucial for validity of MR studies, they should always be discussed in the specific context of the study. If IV methods are used to estimate a causal effect of the exposure, the statistical methods used should be clearly explained. Studies using data from multiple populations should also clearly report how data or estimates are combined.

Supplementary Data

Supplementary data are available at IJE online.

Funding

This work was supported by The Netherlands Organisation for Health Research and Development ZonMw [grant number 152002040].

Acknowledgements

We thank Ruifang Li for helping to extract the required information from an article published in Chinese. All authors contributed to the design of the study. A.G.C.B. performed the literature search and extracted the relevant data. All authors contributed to the interpretation of the data. A.G.C.B. drafted the article and S.l.C. and O.M.D. critically revised it. All authors approved the final version of the article. A.G.C.B. takes responsibility for the integrity of the data, the accuracy of the data analysis and the accuracy and completeness of references.

Conflict of interest: None declared.

References

1
Lawlor
DA
Harbord
RM
Sterne
JA
Timpson
N
Davey Smith
G
.
Mendelian randomization: using genes as instruments for making causal inferences in epidemiology
.
Stat Med
 
2008
;
27
:
1133
63
.
2
Davey Smith
G
Ebrahim
S
.
‘Mendelian randomization': can genetic epidemiology contribute to understanding environmental determinants of disease?
Int J Epidemiol
 
2003
;
32
:
1
22
.
3
Katan
MB
.
Apolipoprotein E isoforms, serum cholesterol, and cancer
.
Lancet
 
1986
;
1
:
507
08
.
4
Taylor
AE
Davies
NM
Ware
JJ
VanderWeele
T
Davey Smith
G
Munafo
MR
.
Mendelian randomization in health research: using appropriate genetic variants and avoiding biased estimates
.
Econ Hum Biol
 
2014
;
13
:
99
106
.
5
Sheehan
NA
Meng
S
Didelez
V
.
Mendelian randomization: a tool for assessing causality in observational epidemiology
.
Methods Mol Biol
 
2011
;
713
:
153
66
.
6
Didelez
V
Meng
S
Sheehan
NA
.
Assumptions of IV methods for observational epidemiology
.
Stat Sci
 
2010
;
25
:
22
40
.
7
Davies
NM
Davey Smith
G
Windmeijer
F
Martin
RM
.
Issues in the reporting and conduct of instrumental variable studies: a systematic review
.
Epidemiology
 
2013
;
24
:
363
69
.
8
Swanson
SA
Hernán
MA
.
Commentary: how to report instrumental variable analyses (suggestions welcome)
.
Epidemiology
 
2013
;
24
:
370
74
.
9
Martens
EP
Pestman
WR
de Boer
A
Belitser
SV
Klungel
OH
.
Instrumental variables: application and limitations
.
Epidemiology
 
2006
;
17
:
260
67
.
10
Sharma
NK
Gupta
A
Prabhakar
S
et al
.
Association between CFH Y402H polymorphism and age related macular degeneration in North Indian cohort
.
PLoS One
 
2013
;
8
:
e70193
.
11
Shaheen
SO
Rutterford
C
Zuccolo
L
et al
.
Prenatal alcohol exposure and childhood atopic disease: a Mendelian randomization approach
.
J Allergy Clin Immunol
 
2014
;
133
:
225
32
.
12
Humphriss
R
Hall
A
May
M
Zuccolo
L
Macleod
J
.
Prenatal alcohol exposure and childhood balance ability: findings from a UK birth cohort study
.
BMJ Open
 
2013
;
3
(
6
).
doi: 10.1136/bmjopen‐2013‐002716
.
13
Bonilla
C
Lawlor
DA
Taylor
AE
et al
.
Vitamin B-12 status during pregnancy and child's IQ at age 8: a Mendelian randomization study in the Avon longitudinal study of parents and children
.
PLoS One
 
2012
;
7
:
e51084
.
14
Bonilla
C
Lawlor
DA
Ben-Shlomo
Y
et al
.
Maternal and offspring fasting glucose and type 2 diabetes-associated genetic variants and cognitive function at age 8: a Mendelian randomization study in the Avon Longitudinal Study of Parents and Children
.
BMC Med Genet
 
2012
;
13
:
90
.
15
Bonilla
C
Gilbert
R
Kemp
JP
et al
.
Using genetic proxies for lifecourse sun exposure to assess the causal relationship of sun exposure with circulating vitamin D and prostate cancer risk
.
Cancer Epidemiol Biomarkers Prev
 
2013
;
22
:
597
606
.
16
Alegret
JM
Aragonès
G
Elosua
R
et al
.
The relevance of the association between inflammation and atrial fibrillation
.
Eur J Clin Invest
 
2013
;
43
:
324
31
.
17
Almon
R
Álvarez-León
EE
Serra-Majem
L
.
Association of the European lactase persistence variant (LCT-13910 C>T polymorphism) with obesity in the Canary Islands
.
PLoS One
 
2012
;
7
:
e43978
.
18
Bjørngaard
JH
Gunnell
D
Elvestad
MB
et al
.
The causal role of smoking in anxiety and depression: a Mendelian randomization analysis of the HUNT study
.
Psychol Med
 
2013
;
43
:
711
19
.
19
Attermann
J
Obel
C
Bilenberg
N
Nordenbæk
CM
Skytthe
A
Olsen
J
.
Traits of ADHD and autism in girls with a twin brother: a Mendelian randomization study
.
Eur Child Adolesc Psychiatry
 
2012
;
21
:
503
09
.
20
Yang
Q
Bailey
L
Clarke
R
et al
.
Prospective study of methylenetetrahydrofolate reductase (MTHFR) variant C677T and risk of all-cause and cardiovascular disease mortality among 6000 US adults
.
Am J Clin Nutr
 
2012
;
95
:
1245
53
.
21
van Durme
YM
Lahousse
L
Verhamme
KM
et al
.
Mendelian randomization study of interleukin-6 in chronic obstructive pulmonary disease
.
Respiration
 
2011
;
82
:
530
38
.
22
Scott
JA
Berkley
JA
Mwangi
I
et al
.
Relation between falciparum malaria and bacteraemia in Kenyan children: a population-based, case-control study and a longitudinal study
.
Lancet
 
2011
;
378
:
1316
23
.
23
Chmielewski
M
Verduijn
M
Drechsler
C
et al
.
Low cholesterol in dialysis patients - causal factor for mortality or an effect of confounding?
Nephrol Dial Transplant
 
2011
;
26
:
3325
31
.
24
Bolton
CE
Schumacher
W
Cockcroft
JR
et al
.
The CRP genotype, serum levels and lung function in men: the Caerphilly Prospective Study
.
Clin Sci (Lond)
 
2011
;
120
:
347
55
.
25
Kröger
J
Zietemann
V
Enzenbach
C
et al
.
Erythrocyte membrane phospholipid fatty acids, desaturase activity, and dietary fatty acids in relation to risk of type 2 diabetes in the European Prospective Investigation into Cancer and Nutrition (EPIC)-Potsdam Study
.
Am J Clin Nutr
 
2011
;
93
:
127
42
.
26
Welsh
P
Polisecki
E
Robertson
M
et al
.
Unraveling the directional link between adiposity and inflammation: a bidirectional Mendelian randomization approach
.
J Clin Endocrinol Metab
 
2010
;
95
:
93
99
.
27
Trompet
S
Jukema
JW
Katan
MB
et al
.
Apolipoprotein e genotype, plasma cholesterol, and cancer: a Mendelian randomization study
.
Am J Epidemiol
 
2009
;
170
:
1415
21
.
28
Almon
R
Álvarez-León
EE
Engfeldt
P
Serra-Majem
L
Magnuson
A
Nilsson
TK
.
Associations between lactase persistence and the metabolic syndrome in a cross-sectional study in the Canary Islands
.
Eur J Nutr
 
2010
;
49
:
141
46
.
29
Brennan
P
McKay
J
Moore
L
et al
.
Obesity and cancer: Mendelian randomization approach utilizing the FTO genotype
.
Int J Epidemiol
 
2009
;
38
:
971
75
.
30
Drenos
F
Talmud
PJ
Casas
JP
et al
.
Integrated associations of genotypes with multiple blood biomarkers linked to coronary heart disease risk
.
Hum Mol Genet
 
2009
;
18
:
2305
16
.
31
Lim
LS
Tai
ES
Aung
T
et al
.
Relation of age-related cataract with obesity and obesity genes in an Asian population
.
Am J Epidemiol
 
2009
;
169
:
1267
74
.
32
Giltay
EJ
van Reedt Dortland
AK
Nissinen
A
et al
.
Serum cholesterol, apolipoprotein E genotype and depressive symptoms in elderly European men: the FINE study
.
J Affect Disord
 
2009
;
115
:
471
77
.
33
Irons
DE
McGue
M
Iacono
WG
Oetting
WS
.
Mendelian randomization: a novel test of the gateway hypothesis and models of gene-environment interplay
.
Dev Psychopathol
 
2007
;
19
:
1181
95
.
34
Herder
C
Klopp
N
Baumert
J
et al
.
Effect of macrophage migration inhibitory factor (MIF) gene variants and MIF serum concentrations on the risk of type 2 diabetes: results from the MONICA/KORA Augsburg Case-Cohort Study, 1984–2002
.
Diabetologia
 
2008
;
51
:
276
84
.
35
Keavney
B
Danesh
J
Parish
S
et al
.
Fibrinogen and coronary heart disease: test of causality by 'Mendelian randomization'
.
Int J Epidemiol
 
2006
;
35
:
935
43
.
36
Bech
BH
Autrup
H
Nohr
EA
Henriksen
TB
Olsen
J
.
Stillbirth and slow metabolizers of caffeine: comparison by genotypes
.
Int J Epidemiol
 
2006
;
35
:
948
53
.
37
Zuccolo
L
Lewis
SJ
Davey Smith
G
et al
.
Prenatal alcohol exposure and offspring cognition and school performance. A ‘Mendelian randomization' natural experiment
.
Int J Epidemiol
 
2013
;
42
:
1358
70
.
38
Aramini
B
Kim
C
Diangelo
S
et al
.
Donor surfactant protein D (SP-D) polymorphisms are associated with lung transplant outcome
.
Am J Transplant
 
2013
;
13
:
2130
36
.
39
Yarwood
A
Martin
P
Bowes
J
et al
.
Enrichment of vitamin D response elements in RA-associated loci supports a role for vitamin D in the pathogenesis of RA
.
Genes Immun
 
2013
;
14
:
325
29
.
40
Travis
RC
Appleby
PN
Siddiq
A
et al
.
Genetic variation in the lactase gene, dairy product intake and risk for prostate cancer in the European prospective investigation into cancer and nutrition
.
Int J Cancer
 
2013
;
132
:
1901
-
–10
.
41
Veen
G
Giltay
EJ
Van Vliet
IM
et al
.
C-reactive protein polymorphisms are associated with the cortisol awakening response in basal conditions in human subjects
.
Stress
 
2011
;
14
:
128
35
.
42
Koshy
B
Miyashita
A
St Jean
P
et al
.
Genetic deficiency of plasma lipoprotein-associated phospholipase A 2 (PLA2G7 V297F null mutation) and risk of Alzheimer's disease in Japan
.
J Alzheimer's Dis
 
2010
;
21
:
775
80
.
43
Pierce
BL
Ahsan
H
.
Genetic susceptibility to type 2 diabetes is associated with reduced prostate cancer risk
.
Hum Hered
 
2010
;
69
:
193
201
.
44
Thuesen
BH
Husemoen
LLN
Fenger
M
Linneberg
A
.
Lack of association between the MTHFR (C677T) polymorphism and atopic disease
.
Clin Resp J
 
2009
;
3
:
102
08
.
45
Granell
R
Heron
J
Lewis
S
Davey Smith
G
Sterne
JAC
Henderson
J
.
The association between mother and child MTHFR C677T polymorphisms, dietary folate intake and childhood atopy in a population-based, longitudinal birth cohort
.
Clin Exp Allergy
 
2008
;
38
:
320
28
.
46
Heidrich
J
Wellmann
J
Doring
A
Illig
T
Keil
U
.
Alcohol consumption, alcohol dehydrogenase and risk of coronary heart disease in the MONICA/KORA-Augsburg cohort 1994/1995–2002
.
Eur J Cardiovasc Prev Rehabil
 
2007
;
14
:
769
74
.
47
Almeida
OP
Hankey
GJ
Yeap
BB
Golledge
J
Flicker
L
.
The triangular association of ADH1B genetic polymorphism, alcohol consumption and the risk of depression in older men
.
Mol Psychiatry
 
2014
;
19
:
995
1000
.
48
Dai
X
Yuan
J
Yao
P
et al
.
Association between serum uric acid and the metabolic syndrome among a middle- and old-age Chinese population
.
Eur J Epidemiol
 
2013
;
28
:
66976
.
49
Yao
WM
Zhang
HF
Zhu
ZY
et al
.
Genetically elevated levels of circulating triglycerides and brachial-ankle pulse wave velocity in a Chinese population
.
J Hum Hypertens
 
2013
;
27
:
265
70
.
50
Gan
W
Guan
Y
Wu
Q
et al
.
Association of TMPRSS6 polymorphisms with ferritin, hemoglobin, and type 2 diabetes risk in a Chinese Han population
.
Am J Clin Nutr
 
2012
;
95
:
626
32
.
51
Breitling
LP
Koenig
W
Fischer
M
et al
.
Type II secretory phospholipase A2 and prognosis in patients with stable coronary heart disease: mendelian randomization study
.
PLoS One
 
2011
;
6
:
e22318
.
52
Rasmussen-Torvik
LJ
Li
M
Kao
WH
et al
.
Association of a fasting glucose genetic risk score with subclinical atherosclerosis: The Atherosclerosis Risk in Communities (ARIC) study
.
Diabetes
 
2011
;
60
:
331
35
.
53
Wu
Y
Li
H
Loos
RJ
et al
.
RBP4 variants are significantly associated with plasma RBP4 levels and hypertriglyceridemia risk in Chinese Hans
.
J Lipid Res
 
2009
;
50
:
1479
86
.
54
Di Paola
R
Marucci
A
Fontana
A
et al
.
Role of obesity on all-cause mortality in whites with type 2 diabetes from Italy
.
Acta Diabetol
 
2013
;
50
:
971
76
.
55
Menzaghi
C
De Cosmo
S
Copetti
M
et al
.
Relationship between ADIPOQ gene, circulating high molecular weight adiponectin and albuminuria in individuals with normal kidney function: Evidence from a family-based study
.
Diabetologia
 
2011
;
54
:
812
18
.
56
Oei
L
Campos-Obando
N
Dehghan
A
et al
.
Dissecting the relationship between high-sensitivity serum C-reactive protein and increased fracture risk: the Rotterdam Study
.
Osteoporos Int
 
2014
;
25
:
1247
54
.
57
Tian
Q
Jia
J
Ling
S
Liu
Y
Yang
S
Shao
Z
.
A causal role for circulating miR-34b in osteosarcoma
.
Eur J Surg Oncol
 
2014
;
40
:
67
72
.
58
Bouthoorn
SH
van Lenthe
FJ
Kiefte-de Jong
JC
et al
.
Genetic taste blindness to bitter and body composition in childhood: a Mendelian randomization design
.
Int J Obes (Lond)
 
2014
;
38
:
1005
10
.
59
Lee
HA
Park
EA
Cho
SJ
et al
.
Mendelian randomization analysis of the effect of maternal homocysteine during pregnancy, as represented by maternal MTHFR C677T genotype, on birth weight
.
J Epidemiol
 
2013
;
23
:
371
75
.
60
Warodomwichit
D
Sritara
C
Thakkinstian
A
et al
.
Causal inference of the effect of adiposity on bone mineral density in adults
.
Clin Endocrinol (Oxf)
 
2013
;
78
:
694
99
.
61
Jensen
MK
Bartz
TM
Djousse
L
et al
.
Genetically elevated fetuin-A levels, fasting glucose levels, and risk of type 2 diabetes: the cardiovascular health study
.
Diabetes Care
 
2013
;
36
:
3121
27
.
62
Gao
H
Fall
T
van Dam
RM
et al
.
Evidence of a causal relationship between adiponectin levels and insulin sensitivity: a Mendelian randomization study
.
Diabetes
 
2013
;
62
:
1338
44
.
63
Alwan
NA
Lawlor
DA
McArdle
HJ
Greenwood
DC
Cade
JE
.
Exploring the relationship between maternal iron status and offspring's blood pressure and adiposity: a Mendelian randomization study
.
Clin Epidemiol
 
2012
;
4
:
193
200
.
64
Oikonen
M
Wendelin-Saarenhovi
M
Lyytikäinen
LP
et al
.
Associations between serum uric acid and markers of subclinical atherosclerosis in young adults. The cardiovascular risk in Young Finns study
.
Atherosclerosis
 
2012
;
223
:
497
503
.
65
Lyngdoh
T
Vuistiner
P
Marques-Vidal
P
et al
.
Serum uric acid and adiposity: deciphering causality using a bidirectional Mendelian randomization approach
.
PLoS One
 
2012
;
7
:
e39321
.
66
Guessous
I
Dobrinas
M
Kutalik
Z
et al
.
Caffeine intake and CYP1A2 variants associated with high caffeine intake protect non-smokers from hypertension
.
Hum Mol Genet
 
2012
;
21
:
3283
92
.
67
Au Yeung
SL
Jiang
C
Cheng
KK
et al
.
Moderate alcohol use and cardiovascular disease from Mendelian randomization
.
PLoS One
 
2013
;
8
:
e68054
.
68
Au Yeung
SL
Jiang
CQ
Cheng
KK
et al
.
Evaluation of moderate alcohol use and cognitive function among men using a Mendelian randomization design in the Guangzhou biobank cohort study
.
Am J Epidemiol
 
2012
;
175
:
1021
28
.
69
Lewis
SJ
Araya
R
Davey Smith
G
et al
.
Smoking is associated with, but does not cause, depressed mood in pregnancy – a mendelian randomization study
.
PLoS One
 
2011
;
6
:
e21689
.
70
Conen
D
Vollenweider
P
Rousson
V
et al
.
Use of a Mendelian randomization approach to assess the causal relation of gamma-Glutamyltransferase with blood pressure and serum insulin levels
.
Am J Epidemiol
 
2010
;
172
:
1431
41
.
71
Bochud
M
Marquant
F
Marques-Vidal
PM
et al
.
Association between C-reactive protein and adiposity in women
.
J Clin Endocrinol Metab
 
2009
;
94
:
3969
77
.
72
Timpson
NJ
Sayers
A
Davey Smith
G
Tobias
JH
.
How does body fat influence bone mass in childhood? A Mendelian randomization approach
.
J Bone Miner Res
 
2009
;
24
:
522
33
.
73
Kivimäki
M
Lawlor
DA
Davey Smith
G
et al
.
Does high C-reactive protein concentration increase atherosclerosis? The Whitehall II Study
.
PLoS One
 
2008
;
3
:
e3013
.
74
Brunner
EJ
Kivimäki
M
Witte
DR
et al
.
Inflammation, insulin resistance, and diabetes – Mendelian randomization using CRP haplotypes points upstream
.
PLoS Med
 
2008
;
5
:
e155
.
75
Lawlor
DA
Timpson
NJ
Harbord
RM
et al
.
Exploring the developmental overnutrition hypothesis using parental-offspring associations and FTO as an instrumental variable
.
PLoS Med
 
2008
;
5
:
e33
.
76
Viikari
LA
Huupponen
RK
Viikari
JS
et al
.
Relationship between leptin and C-reactive protein in young Finnish adults
.
J Clin Endocrinol Metab
 
2007
;
92
:
4753
58
.
77
Timpson
NJ
Lawlor
DA
Harbord
RM
et al
.
C-reactive protein and its role in metabolic syndrome: mendelian randomization study
.
Lancet
 
2005
;
366
:
1954
59
.
78
Binder
AM
Michels
KB
.
The causal effect of red blood cell folate on genome-wide methylation in cord blood: a Mendelian randomization approach
.
BMC Bioinformatics
 
2013
;
14
:
353
.
79
Debette
S
Wolf
C
Lambert
JC
et al
.
Abdominal obesity and lower gray matter volume: a Mendelian randomization study
.
Neurobiol Aging
 
2014
;
35
:
378
86
.
80
Thakkinstian
A
Chailurkit
L
Warodomwichit
D
et al
.
Causal relationship between body mass index and fetuin-A level in the asian population: a bidirectional mendelian randomization study
.
Clin Endocrinol (Oxf)
 
2014
;
81
:
197
203
.
81
Haring
R
Teumer
A
Völker
U
et al
.
Mendelian randomization suggests non-causal associations of testosterone with cardiometabolic risk factors and mortality
.
Andrology
 
2013
;
1
:
17
23
.
82
Islam
M
Jafar
TH
Wood
AR
et al
.
Multiple genetic variants explain measurable variance in type 2 diabetes-related traits in Pakistanis
.
Diabetologia
 
2012
;
55
:
2193
204
.
83
Kivimäki
M
Jokela
M
Hamer
M
et al
.
Examining overweight and obesity as risk factors for common mental disorders using fat mass and obesity-associated (FTO) genotype-instrumented analysis: The Whitehall II Study, 1985–2004
.
Am J Epidemiol
 
2011
;
173
:
421
29
.
84
Jokela
M
Elovainio
M
Keltikangas-Järvinen
L
et al
.
Body mass index and depressive symptoms: instrumental-variables regression with genetic risk score
.
Genes Brain Behav
 
2012
;
Sep 7. doi: 10.1111/j.1601‐183X.2012.00846.x. [Epub ahead of print.]
85
Kivimäki
M
Magnussen
CG
Juonala
M
et al
.
Conventional and Mendelian randomization analyses suggest no association between lipoprotein(a) and early atherosclerosis: the Young Finns Study
.
Int J Epidemiol
 
2011
;
40
:
470
78
.
86
Kivimäki
M
Davey Smith
G
Timpson
NJ
et al
.
Lifetime body mass index and later atherosclerosis risk in young adults: examining causal links using Mendelian randomization in the Cardiovascular Risk in Young Finns study
.
Eur Heart J
 
2008
;
29
:
2552
60
.
87
Frayling
TM
Rafiq
S
Murray
A
et al
.
An interleukin-18 polymorphism is associated with reduced serum concentrations and better physical functioning in older people
.
J Gerontol A Biol Sci Med Sci
 
2007
;
62
:
73
78
.
88
Davey Smith
G
Lawlor
DA
Harbord
R
et al
.
Association of C-reactive protein with blood pressure and hypertension: life course confounding and mendelian randomization tests of causality
.
Arterioscler Thromb Vasc Biol
 
2005
;
25
:
1051
56
.
89
Lawlor
DA
Nordestgaard
BG
Benn
M
Zuccolo
L
Tybjærg-Hansen
A
Davey Smith
G
.
Exploring causal associations between alcohol and coronary heart disease risk factors: findings from a Mendelian randomization study in the Copenhagen General Population Study
.
Eur Heart J
 
2013
;
34
:
2519
28
.
90
Theodoratou
E
Palmer
T
Zgaga
L
et al
.
Instrumental variable estimation of the causal effect of plasma 25-hydroxy-vitamin D on colorectal cancer risk: a mendelian randomization analysis
.
PLoS One
 
2012
;
7
:
e37662
.
91
Collin
SM
Metcalfe
C
Palmer
TM
et al
.
The causal roles of vitamin B(12) and transcobalamin in prostate cancer: can Mendelian randomization analysis provide definitive answers?
Int J Mol Epidemiol Genet
 
2011
;
2
:
316
27
.
92
Lawlor
DA
Harbord
RM
Tybjærg-Hansen
A
et al
.
Using genetic loci to understand the relationship between adiposity and psychological distress: a Mendelian Randomization study in the Copenhagen General Population Study of 53 221 adults
.
J Intern Med
 
2011
;
269
:
525
37
.
93
Mumby
HS
Elks
CE
Li
S
et al
.
Mendelian randomization study of childhood BMI and early menarche
.
J Obes
 
2011
;
2011
:
180729
.
94
Timpson
NJ
Nordestgaard
BG
Harbord
RM
et al
.
C-reactive protein levels and body mass index: elucidating direction of causation through reciprocal Mendelian randomization
.
Int J Obes (Lond)
 
2011
;
35
:
300
08
.
95
Zacho
J
Tybjærg-Hansen
A
Nordestgaard
BG
.
C-reactive protein and all-cause mortality - the Copenhagen City Heart Study
.
Eur Heart J
 
2010
;
31
:
1624
32
.
96
Verduijn
M
Prein
RA
Stenvinkel
P
et al
.
Is fetuin-A a mortality risk factor in dialysis patients or a mere risk marker? A Mendelian randomization approach
.
Nephrol Dial Transplant
 
2011
;
26
:
239
45
.
97
Fisher
E
Stefan
N
Saar
K
et al
.
Association of AHSG gene polymorphisms with fetuin-A plasma levels and cardiovascular diseases in the EPIC-Potsdam study
.
Circ Cardiovasc Genet
 
2009
;
2
:
607
13
.
98
Timpson
NJ
Harbord
R
Davey Smith
G
Zacho
J
Tybjærg-Hansen
A
Nordestgaard
BG
.
Does greater adiposity increase blood pressure and hypertension risk?: Mendelian randomization using the FTO/MC4R genotype
.
Hypertension
 
2009
;
54
:
84
90
.
99
Kang
H
Kreuels
B
Adjei
O
Krumkamp
R
May
J
Small
DS
.
The causal effect of malaria on stunting: a Mendelian randomization and matching approach
.
Int J Epidemiol
 
2013
;
42
:
1390
98
.
100
Qin
XY
Tian
J
Fang
K
et al
.
[Mendelian randomization study of the relationship between high-density lipoprotein cholesterol and age-related macular degeneration]
.
Beijing Da Xue Xue Bao
 
2012
;
44
:
407
11
.
101
Trummer
O
Pilz
S
Hoffmann
MM
et al
.
Vitamin D and mortality: a Mendelian randomization study
.
Clin Chem
 
2013
;
59
:
793
97
.
102
You
NC
Chen
BH
Song
Y
et al
.
A prospective study of leukocyte telomere length and risk of type 2 diabetes in postmenopausal women
.
Diabetes
 
2012
;
61
:
2998
3004
.
103
McArdle
PF
Whitcomb
BW
Tanner
K
Mitchell
BD
Shuldiner
AR
Parsa
A
.
Association between bilirubin and cardiovascular disease risk factors: using Mendelian randomization to assess causal inference
.
BMC Cardiovasc Disord
 
2012
;
12
:
16
.
104
Parsa
A
Brown
E
Weir
MR
et al
.
Genotype-based changes in serum uric acid affect blood pressure
.
Kidney Int
 
2012
;
81
:
502
07
.
105
Sunyer
J
Pistelli
R
Plana
E
et al
.
Systemic inflammation, genetic susceptibility and lung function
.
Eur Respir J
 
2008
;
32
:
92
97
.
106
Mente
A
Meyre
D
Lanktree
MB
et al
.
Causal relationship between adiponectin and metabolic traits: a Mendelian randomization study in a multiethnic population
.
PLoS One
 
2013
;
8
:
e66808
.
107
Love-Gregory
L
Sherva
R
Schappe
T
et al
.
Common CD36 SNPs reduce protein expression and may contribute to a protective atherogenic profile
.
Hum Mol Genet
 
2011
;
20
:
193
201
.
108
Nagele
P
Zeugswetter
B
Wiener
C
et al
.
Influence of methylenetetrahydrofolate reductase gene polymorphisms on homocysteine concentrations after nitrous oxide anesthesia
.
Anesthesiology
 
2008
;
109
:
36
43
.
109
Klovaite
J
Nordestgaard
BG
Tybjærg-Hansen
A
Benn
M
.
Elevated fibrinogen levels are associated with risk of pulmonary embolism, but not with deep venous thrombosis
.
Am J Respir Crit Care Med
 
2013
;
187
:
286
93
.
110
Rius-Ottenheim
N
de Craen
AJ
Geleijnse
JM
et al
.
C-reactive protein haplotypes and dispositional optimism in obese and nonobese elderly subjects
.
Inflamm Res
 
2012
;
61
:
43
51
.
111
Stender
S
Frikke-Schmidt
R
Nordestgaard
BG
Tybjærg-Hansen
A
.
Extreme bilirubin levels as a causal risk factor for symptomatic gallstone disease
.
JAMA Intern Med
 
2013
;
173
:
1222
28
.
112
Dahl
M
Vestbo
J
Zacho
J
Lange
P
Tybjærg-Hansen
A
Nordestgaard
BG
.
C reactive protein and chronic obstructive pulmonary disease: a Mendelian randomization approach
.
Thorax
 
2011
;
66
:
197
204
.
113
Marott
SC
Nordestgaard
BG
Zacho
J
et al
.
Does elevated C-reactive protein increase atrial fibrillation risk? A Mendelian randomization of 47 000 individuals from the general population
.
J Am Coll Cardiol
 
2010
;
56
:
789
95
.
114
Skaaby
T
Husemoen
LL
Martinussen
T
et al
.
Vitamin D status, filaggrin genotype, and cardiovascular risk factors: a Mendelian randomization approach
.
PLoS One
 
2013
;
8
:
e57647
.
115
Cruchaga
C
Kauwe
JS
Nowotny
P
et al
.
Cerebrospinal fluid APOE levels: an endophenotype for genetic studies for Alzheimer's disease
.
Hum Mol Genet
 
2012
;
21
:
4558
71
.
116
Hughes
K
Flynn
T
de Zoysa
J
Dalbeth
N
Merriman
TR
.
Mendelian randomization analysis associates increased serum urate, due to genetic variation in uric acid transporters, with improved renal function
.
Kidney Int
 
2014
;
85
:
344
51
.
117
Benn
M
Tybjærg-Hansen
A
McCarthy
MI
Jensen
GB
Grande
P
Nordestgaard
BG
.
Nonfasting glucose, ischemic heart disease, and myocardial infarction: a Mendelian randomization study
.
J Am Coll Cardiol
 
2012
;
59
:
2356
65
.
118
Varbo
A
Benn
M
Tybjærg-Hansen
A
Nordestgaard
BG
.
Elevated remnant cholesterol causes both low-grade inflammation and ischemic heart disease, whereas elevated low-density lipoprotein cholesterol causes ischemic heart disease without inflammation
.
Circulation
 
2013
;
128
:
1298
309
.
119
Stender
S
Nordestgaard
BG
Tybjærg-Hansen
A
.
Elevated body mass index as a causal risk factor for symptomatic gallstone disease: a Mendelian randomization study
.
Hepatology
 
2013
;
58
:
2133
41
.
120
Kamstrup
PR
Nordestgaard
BG
.
Lipoprotein(a) concentrations, isoform size, and risk of type 2 diabetes: a Mendelian randomization study
.
Lancet Diabet Endocrinol
 
2013
;
1
:
220
27
.
121
Wium-Andersen
MK
Orsted
DD
Nordestgaard
BG
.
Elevated C-reactive protein associated with late- and very-late-onset schizophrenia in the general population: a prospective study
.
Schizophr Bull
 
2014
;
40
:
1117
27
.
122
Heikkilä
K
Silander
K
Salomaa
V
et al
.
C-reactive protein-associated genetic variants and cancer risk: findings from FINRISK 1992, FINRISK 1997 and Health 2000 studies
.
Eur J Cancer
 
2011
;
47
:
404
12
.
123
Allin
KH
Nordestgaard
BG
Zacho
J
Tybjærg-Hansen
A
Bojesen
SE
.
C-reactive protein and the risk of cancer: a mendelian randomization study
.
J Natl Cancer Inst
 
2010
;
102
:
202
06
.
124
Kamstrup
PR
Tybjærg-Hansen
A
Nordestgaard
BG
.
Elevated lipoprotein(a) and risk of aortic valve stenosis in the general population
.
J Am Coll Cardiol
 
2014
;
63
:
470
77
.
125
Pierce
BL
Tong
L
Argos
M
et al
.
Arsenic metabolism efficiency has a causal role in arsenic toxicity: Mendelian randomization and gene-environment interaction
.
Int J Epidemiol
 
2013
;
42
:
1862
71
.
126
Oelsner
EC
Pottinger
TD
Burkart
KM
et al
.
Adhesion molecules, endothelin-1 and lung function in seven population-based cohorts
.
Biomarkers
 
2013
;
18
:
196
203
.
127
Trombetta
M
Bonetti
S
Boselli
ML
et al
.
PPARG2 Pro12Ala and ADAMTS9 rs4607103 as ‘insulin resistance loci’ and ‘insulin secretion loci’ in Italian individuals. The GENFIEV study and the Verona Newly Diagnosed Type 2 Diabetes Study (VNDS) 4
.
Acta Diabetol
 
2013
;
50
:
401
08
.
128
Varbo
A
Benn
M
Tybjærg-Hansen
A
Jørgensen
AB
Frikke-Schmidt
R
Nordestgaard
BG
.
Remnant cholesterol as a causal risk factor for ischemic heart disease
.
J Am Coll Cardiol
 
2013
;
61
:
427
36
.
129
Jørgensen
AB
Frikke-Schmidt
R
West
AS
Grande
P
Nordestgaard
BG
Tybjærg-Hansen
A
.
Genetically elevated non-fasting triglycerides and calculated remnant cholesterol as causal risk factors for myocardial infarction
.
Eur Heart J
 
2013
;
34
:
1826
33
.
130
Haase
CL
Tybjærg-Hansen
A
Qayyum
AA
Schou
J
Nordestgaard
BG
Frikke-Schmidt
R
.
LCAT, HDL cholesterol and ischemic cardiovascular disease: a Mendelian randomization study of HDL cholesterol in 54 500 individuals
.
J Clin Endocrinol Metab
 
2012
;
97
:
E248
E256
.
131
Benn
M
Tybjærg-Hansen
A
Stender
S
Frikke-Schmidt
R
Nordestgaard
BG
.
Low-density lipoprotein cholesterol and the risk of cancer: a mendelian randomization study
.
J Natl Cancer Inst
 
2011
;
103
:
508
19
.
132
Zacho
J
Tybjærg-Hansen
A
Nordestgaard
BG
.
C-reactive protein and risk of venous thromboembolism in the general population
.
Arterioscler Thromb Vasc Biol
 
2010
;
30
:
1672
78
.
133
Wium-Andersen
MK
Orsted
DD
Nordestgaard
BG
.
Elevated C-reactive protein, depression, somatic diseases, and all-cause mortality: a mendelian randomization study
.
Biol Psychiatry
 
2014
;
76
:
249
57
.
134
Stender
S
Frikke-Schmidt
R
Benn
M
Nordestgaard
BG
Tybjærg-Hansen
A
.
Low-density lipoprotein cholesterol and risk of gallstone disease: a Mendelian randomization study and meta-analyses
.
J Hepatol
 
2013
;
58
:
126
33
.
135
Yaghootkar
H
Lamina
C
Scott
RA
et al
.
Mendelian randomization studies do not support a causal role for reduced circulating adiponectin levels in insulin resistance and type 2 diabetes
.
Diabetes
 
2013
;
62
:
3589
98
.
136
Chen
L
Davey Smith
G
Harbord
RM
Lewis
SJ
.
Alcohol intake and blood pressure: a systematic review implementing a Mendelian randomization approach
.
PLoS Med
 
2008
;
5
:
e52
.
137
Shah
S
Casas
JP
Drenos
F
et al
.
Causal relevance of blood lipid fractions in the development of carotid atherosclerosis: Mendelian randomization analysis
.
Circ Cardiovasc Genet
 
2013
;
6
:
63
72
.
138
Palmer
TM
Nordestgaard
BG
Benn
M
et al
.
Association of plasma uric acid with ischaemic heart disease and blood pressure: mendelian randomization analysis of two large cohorts
.
BMJ
 
2013
;
347
:
f4262
.
139
De Silva
NM
Freathy
RM
Palmer
TM
et al
.
Mendelian randomization studies do not support a role for raised circulating triglyceride levels influencing type 2 diabetes, glucose levels, or insulin resistance
.
Diabetes
 
2011
;
60
:
1008
18
.
140
Thanassoulis
G
Campbell
CY
Owens
DS
et al
.
Genetic associations with valvular calcification and aortic stenosis
.
N Engl J Med
 
2013
;
368
:
503
12
.
141
Nordestgaard
BG
Palmer
TM
Benn
M
et al
.
The effect of elevated body mass index on ischemic heart disease risk: causal estimates from a Mendelian randomization approach
.
PLoS Med
 
2012
;
9
:
e1001212
.
142
Elliott
P
Chambers
JC
Zhang
W
et al
.
Genetic loci associated with C-reactive protein levels and risk of coronary heart disease
.
JAMA
 
2009
;
302
:
37
48
.
143
Ye
Z
Haycock
PC
Gurdasani
D
et al
.
The association between circulating lipoprotein(a) and type 2 diabetes: is it causal?
Diabetes
 
2014
;
63
:
332
42
.
144
Voight
BF
Peloso
GM
Orho-Melander
M
et al
.
Plasma HDL cholesterol and risk of myocardial infarction: a mendelian randomization study
.
Lancet
 
2012
;
380
:
572
80
.
145
Mamasoula
C
Prentice
RR
Pierscionek
et al
.
Association between C677T polymorphism of methylene tetrahydrofolate reductase and congenital heart disease: meta-analysis of 7697 cases and 13 125 controls
.
Circ Cardiovasc Genet
 
2013
;
6
:
347
53
.
146
Harrison
SC
Smith
AJ
Jones
GT
et al
.
Interleukin-6 receptor pathways in abdominal aortic aneurysm
.
Eur Heart J
 
2013
;
34
:
3707
16
.
147
Stender
S
Frikke-Schmidt
R
Nordestgaard
BG
Grande
P
Tybjærg-Hansen
A
.
Genetically elevated bilirubin and risk of ischaemic heart disease: three Mendelian randomization studies and a meta-analysis
.
J Intern Med
 
2013
;
273
:
59
68
.
148
Hingorani
AD
Casas
JP
.
The interleukin-6 receptor as a target for prevention of coronary heart disease: a mendelian randomization analysis
.
Lancet
 
2012
;
379
:
1214
24
.
149
Clarke
R
Bennett
DA
Parish
S
et al
.
Homocysteine and coronary heart disease: meta-analysis of MTHFR case-control studies, avoiding publication bias
.
PLoS Med
 
2012
;
9
:
e1001177
.
150
Wang
J
Wang
H
Chen
Y
Hao
P
Zhang
Y
.
Alcohol ingestion and colorectal neoplasia: a meta-analysis based on a Mendelian randomization approach
.
Colorectal Dis
 
2011
;
13
:
e71
e78
.
151
Casas
JP
Ninio
E
Panayiotou
A
et al
.
PLA2G7 genotype, lipoprotein-associated phospholipase A2 activity, and coronary heart disease risk in 10 494 cases and 15 624 controls of European ancestry
.
Circulation
 
2010
;
121
:
2284
93
.
152
Lewis
SJ
Baker
I
Davey Smith
G
.
Meta-analysis of vitamin D receptor polymorphisms and pulmonary tuberculosis risk
.
Int J Tuberc Lung Dis
 
2005
;
9
:
1174
77
.
153
Davies
JR
Field
S
Randerson-Moor
J
et al
.
An inherited variant in the gene coding for vitamin D-binding protein and survival from cutaneous melanoma: a BioGenoMEL study
.
Pigment Cell Melanoma Res
 
2014
;
27
:
234
43
.
154
Panoutsopoulou
K
Metrustry
S
Doherty
SA
et al
.
The effect of FTO variation on increased osteoarthritis risk is mediated through body mass index: a mendelian randomization study
.
Ann Rheum Dis
 
2013
;
73
:
2082
86
.
155
Ioannidis
A
Ikonomi
E
Dimou
NL
Douma
L
Bagos
PG
.
Polymorphisms of the insulin receptor and the insulin receptor substrates genes in polycystic ovary syndrome: a Mendelian randomization meta-analysis
.
Mol Genet Metab
 
2010
;
99
:
174
83
.
156
Rice
NE
Bandinelli
S
Corsi
AM
et al
.
The paraoxonase (PON1) Q192R polymorphism is not associated with poor health status or depression in the ELSA or INCHIANTI studies
.
Int J Epidemiol
 
2009
;
38
:
1374
79
.
157
Rafiq
S
Melzer
D
Weedon
MN
et al
.
Gene variants influencing measures of inflammation or predisposing to autoimmune and inflammatory diseases are not associated with the risk of type 2 diabetes
.
Diabetologia
 
2008
;
51
:
2205
13
.
158
Pfister
R
Sharp
S
Luben
R
et al
.
Mendelian randomization study of B-type natriuretic peptide and type 2 diabetes: evidence of causal association from population studies
.
PLoS Med
 
2011
;
8
:
e1001112
.
159
Pfister
R
Barnes
D
Luben
R
et al
.
No evidence for a causal link between uric acid and type 2 diabetes: a Mendelian randomization approach
.
Diabetologia
 
2011
;
54
:
2561
69
.
160
Marjot
T
Yadav
S
Hasan
N
Bentley
P
Sharma
P
.
Genes associated with adult cerebral venous thrombosis
.
Stroke
 
2011
;
42
:
913
18
.
161
Sarwar
N
Sandhu
MS
Ricketts
SL
et al
.
Triglyceride-mediated pathways and coronary disease: collaborative analysis of 101 studies
.
Lancet
 
2010
;
375
:
1634
39
.
162
Bentley
P
Peck
G
Smeeth
L
Whittaker
J
Sharma
P
.
Causal relationship of susceptibility genes to ischemic stroke: comparison to ischemic heart disease and biochemical determinants
.
PLoS One
 
2010
;
5
:
e9136
.
163
Perry
JR
Ferrucci
L
Bandinelli
S
et al
.
Circulating beta-carotene levels and type 2 diabetes – cause or effect?
Diabetologia
 
2009
;
52
:
2117
21
.
164
Perry
JR
Weedon
MN
Langenberg
C
et al
.
Genetic evidence that raised sex hormone binding globulin (SHBG) levels reduce the risk of type 2 diabetes
.
Hum Mol Genet
 
2010
;
19
:
535
44
.
165
Boccia
S
Hashibe
M
Galli
P
et al
.
Aldehyde dehydrogenase 2 and head and neck cancer: a meta-analysis implementing a Mendelian randomization approach
.
Cancer Epidemiol Biomarkers Prev
 
2009
;
18
:
248
54
.
166
Casas
JP
Bautista
LE
Smeeth
L
Sharma
P
Hingorani
AD
.
Homocysteine and stroke: evidence on a causal link from mendelian randomization
.
Lancet
 
2005
;
365
:
224
32
.
167
Casas
JP
Shah
T
Cooper
J
et al
.
Insight into the nature of the CRP-coronary event association using Mendelian randomization
.
Int J Epidemiol
 
2006
;
35
:
922
31
.
168
Fall
T
Hagg
S
Magi
R
et al
.
The role of adiposity in cardiometabolic traits: a Mendelian randomization analysis
.
PLoS Med
 
2013
;
10
:
e1001474
.
169
Pichler
I
Del Greco
MF
Gögele
M
et al
.
Serum iron levels and the risk of Parkinson disease: a Mendelian randomization study
.
PLoS Med
 
2013
;
10
:
e1001462
.
170
Vimaleswaran
KS
Berry
DJ
Lu
C
et al
.
Causal relationship between obesity and vitamin D status: bi-directional Mendelian randomization analysis of multiple cohorts
.
PLoS Med
 
2013
;
10
:
e1001383
.
171
Ference
BA
Yoo
W
Alesh
I
et al
.
Effect of long-term exposure to lower low-density lipoprotein cholesterol beginning early in life on the risk of coronary heart disease: a Mendelian randomization analysis
.
J Am Coll Cardiol
 
2012
;
60
:
2631
39
.
172
Niu
W
Liu
Y
Qi
Y
Wu
Z
Zhu
D
Jin
W
.
Association of interleukin-6 circulating levels with coronary artery disease: a meta-analysis implementing mendelian randomization approach
.
Int J Cardiol
 
2012
;
157
:
243
52
.
173
Niu
W
Zhang
X
Qi
Y
.
Association of an apolipoprotein E polymorphism with circulating cholesterols and hypertension: a meta-based Mendelian randomization analysis
.
Hypertens Res
 
2012
;
35
:
434
40
.
174
Huang
T
Ren
J
Huang
J
Li
D
.
Association of homocysteine with type 2 diabetes: a meta-analysis implementing Mendelian randomization approach
.
BMC Genomics
 
2013
;
14
:
867
.
175
Davey Smith
G
Harbord
R
Milton
J
Ebrahim
S
Sterne
JAC
.
Does elevated plasma fibrinogen increase the risk of coronary heart disease? Evidence from a meta-analysis of genetic association studies
.
Arterioscler Thromb Vasc Biol
 
2005
;
25
:
2228
33
.
176
Lawlor
DA
Harbord
RM
Timpson
NJ
et al
.
The association of C-reactive protein and CRP genotype with coronary heart disease: findings from five studies with 4 610 cases amongst 18 637 participants
.
PLoS One
 
2008
;
3
:
e3011
.
177
Kamstrup
PR
Tybjærg-Hansen
A
Steffensen
R
Nordestgaard
BG
.
Genetically elevated lipoprotein(a) and increased risk of myocardial infarction
.
JAMA
 
2009
;
301
:
2331
39
.
178
Kamstrup
PR
Tybjærg-Hansen
A
Nordestgaard
BG
.
Genetic evidence that lipoprotein(a) associates with atherosclerotic stenosis rather than venous thrombosis
.
Arterioscler Thromb Vasc Biol
 
2012
;
32
:
1732
41
.
179
Stegeman
BH
Helmerhorst
FM
Vos
HL
Rosendaal
FR
Van Hylckama Vlieg
A
.
Sex hormone-binding globulin levels are not causally related to venous thrombosis risk in women not using hormonal contraceptives
.
J Thromb Haemostat
 
2012
;
10
:
2061
67
.
180
Adamsson Eryd
S
Sjögren
M
Smith
JG
et al
.
Ceruloplasmin and atrial fibrillation: evidence of causality from a population-based Mendelian randomization study
.
J Intern Med
 
2014
;
275
:
164
71
.
181
Ding
EL
Song
Y
Manson
JE
et al
.
Sex hormone-binding globulin and risk of type 2 diabetes in women and men
.
N Engl J Med
 
2009
;
361
:
1152
63
.
182
Husemoen
LL
Skaaby
T
Martinussen
T
et al
.
Investigating the causal effect of vitamin D on serum adiponectin using a mendelian randomization approach
.
Eur J Clin Nutr
 
2014
;
68
:
189
95
.
183
Song
Y
Yeung
E
Liu
A
et al
.
Pancreatic beta-cell function and type 2 diabetes risk: quantify the causal effect using a Mendelian randomization approach based on meta-analyses
.
Hum Mol Genet
 
2012
;
21
:
5010
18
.
184
Wensley
F
Gao
P
Burgess
S
et al
.
Association between C reactive protein and coronary heart disease: mendelian randomization analysis based on individual participant data
.
BMJ
 
2011
;
342
:
d548
.
185
Zhao
J
Jiang
C
Lam
TH
et al
.
Genetically predicted testosterone and cardiovascular risk factors in men: a Mendelian randomization analysis in the Guangzhou Biobank Cohort Study
.
Int J Epidemiol
 
2014
;
43
:
140
48
.
186
Holmes
MV
Simon
T
Exeter
HJ
et al
.
Secretory phospholipase A(2)-IIA and cardiovascular disease: a mendelian randomization study
.
J Am Coll Cardiol
 
2013
;
62
:
1966
76
.
187
Linsel-Nitschke
P
Götz
A
Erdmann
J
et al
.
Lifelong reduction of LDL-cholesterol related to a common variant in the LDL-receptor gene decreases the risk of coronary artery disease – a Mendelian Randomization study
.
PLoS One
 
2008
;
3
:
e2986
.
188
Kunutsor
SK
Burgess
S
Munroe
PB
Khan
H
.
Vitamin D and high blood pressure: causal association or epiphenomenon?
Eur J Epidemiol
 
2014
;
29
:
1
14
.
189
Bochud
M
Rousson
V
.
Usefulness of Mendelian randomization in observational epidemiology
.
Int J Environ Res Public Health
 
2010
;
7
:
711
28
.
190
VanderWeele
TJ
Tchetgen Tchetgen
EJ
Cornelis
M
Kraft
P
.
Methodological challenges in mendelian randomization
.
Epidemiology
 
2014
;
25
:
427
35
.
191
Boef
AG
Dekkers
OM
le Cessie
S
Vandenbroucke
JP
.
Reporting instrumental variable analyses
.
Epidemiology
 
2013
;
24
:
937
38
.
192
Davey Smith
G
Hemani
G
.
Mendelian randomization: genetic anchors for causal inference in epidemiological studies
.
Hum Mol Genet
 
2014
;
23
:
R89
R98
.
193
Burgess
S
Thompson
SG
.
Use of allele scores as instrumental variables for Mendelian randomization
.
Int J Epidemiol
 
2013
;
42
:
1134
44
.
194
Palmer
TM
Sterne
JA
Harbord
RM
et al
.
Instrumental variable estimation of causal risk ratios and causal odds ratios in Mendelian randomization analyses
.
Am J Epidemiol
 
2011
;
173
:
1392
403
.
195
Rassen
JA
Schneeweiss
S
Glynn
RJ
Mittleman
MA
Brookhart
MA
.
Instrumental variable analysis for estimation of treatment effects with dichotomous outcomes
.
Am J Epidemiol
 
2009
;
169
:
273
84
.
196
Vuistiner
P
Bochud
M
Rousson
V
.
A comparison of three methods of Mendelian randomization when the genetic instrument, the risk factor and the outcome are all binary
.
PLoS One
 
2012
;
7
:
e35951
.
197
Burgess
S
.
Identifying the odds ratio estimated by a two-stage instrumental variable analysis with a logistic regression model
.
Stat Med
 
2013
;
32
:
4726
47
.
198
Clarke
PS
Windmeijer
F
.
Instrumental variable estimators for binary outcomes
.
J Am Stat Assoc
 
2012
;
107
:
1638
52
.

Supplementary data