CoChat
Sign Up
Cited by 543Advanced Causal Inference TechniquesGenetic Associations and EpidemiologyStatistical Methods in Epidemiology

Network Mendelian randomization: using genetic variants as instrumental variables to investigate mediation in causal pathways.

Open access
Burgess, S., Daniel, R. M., Butterworth, A. S., Thompson, S. G., the EPIC-InterAct Consortium·2015·International Journal of Epidemiology
DOI 10.1093/ije/dyu176PMC PMC4469795
Open this paper in CoChat — free accountOpen PDF

Abstract

BACKGROUND: Mendelian randomization uses genetic variants, assumed to be instrumental variables for a particular exposure, to estimate the causal effect of that exposure on an outcome. If the instrumental variable criteria are satisfied, the resulting estimator is consistent even in the presence of unmeasured confounding and reverse causation. METHODS: We extend the Mendelian randomization paradigm to investigate more complex networks of relationships between variables, in particular where some of the effect of an exposure on the outcome may operate through an intermediate variable (a mediator). If instrumental variables for the exposure and mediator are available, direct and indirect effects of the exposure on the outcome can be estimated, for example using either a regression-based method or structural equation models. The direction of effect between the exposure and a possible mediator can also be assessed. Methods are illustrated in an applied example considering causal relationships between body mass index, C-reactive protein and uric acid. RESULTS: These estimators are consistent in the presence of unmeasured confounding if, in addition to the instrumental variable assumptions, the effects of both the exposure on the mediator and the mediator on the outcome are homogeneous across individuals and linear without interactions. Nevertheless, a simulation study demonstrates that even considerable heterogeneity in these effects does not lead to bias in the estimates. CONCLUSIONS: These methods can be used to estimate direct and indirect causal effects in a mediation setting, and have potential for the investigation of more complex networks between multiple interrelated exposures and disease outcomes.

AI snapshot

Written from the paper’s full text.

TL;DR

Burgess extends Mendelian randomization to mediation networks, using genetic instrumental variables for an exposure and a mediator to estimate direct and indirect causal effects in the presence of unmeasured confounding. The proposed regression-based and structural equation model estimators perform well in simulations and are illustrated with BMI, C-reactive protein, and uric acid data.

Key findings

  • Direct and indirect effects can be estimated when IVs are available for both exposure and mediator, under IV assumptions plus linearity without interactions and homogeneity of individual-level effects.
  • In simulations with 5000 individuals, 1000 datasets per scenario, and 12 scenarios (12 000 datasets total), both regression-based and SEM methods estimated average direct and indirect effects without substantial bias even with heterogeneity τ² = 0.22 and 0.42; SEM estimates were more efficient, with lower mean standard errors.
  • A causal effect of X on Z was detected in all but seven of 12 000 datasets (>99.9%), while a causal effect of Z on X was found in 4.8%, consistent with chance.
  • In the applied example using 7158 EPIC-InterAct participants, the BMI allele score was associated with CRP (P = 0.009) but the CRP allele score was not clearly associated with BMI (P = 0.17); the direct effect of BMI on uric acid not via CRP was 0.053 (SE 0.035), total effect 0.052 (0.032), and indirect effect −0.001 (0.016), suggesting no mediation via CRP.

Methods

  • The paper develops regression-based approaches—ratio/Wald method, two-stage least squares, and multiple-stage least squares—and a structural equation modelling approach to estimate direct and indirect effects using separate IVs for exposure and mediator.
  • A simulation study generated data under linear models with varying heterogeneity in individual-level causal effects (τ² = 0, 0.22, 0.42), plus sensitivity analyses for interaction and correlated effects.
  • An applied example used data on BMI, CRP, and uric acid from 7158 participants in the EPIC-InterAct study, with weighted allele scores as IVs and adjustment for age, sex, and centre.

Limitations

  • Stated by the authors: consistency formally requires IV assumptions, linearity without interactions, and homogeneity of the exposure–mediator and mediator–outcome effects; simulation showed some bias when an interaction term had non-zero mean and slight bias in the direct effect when individual-level effects were correlated with substantial heterogeneity.
  • Stated by the authors: methods assume continuous variables and do not address binary exposures, mediators, or outcomes with non-collapsibility; post-treatment confounders cannot in general be identified without further assumptions; pleiotropy can make mediation assessment problematic; Mendelian randomization cannot address time-varying effects or feedback.
  • Stated by the authors: SEM estimation relies on multivariate normality and correct path-diagram specification; the applied example is observational and uses allele scores, so causal interpretation depends on the validity of the IV assumptions.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1093/ije/dyu176

Cites

66 of 66 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 35 cited sources checked · 0 supported · 0 flagged

  • Findingown data

    A simulation study demonstrates that even considerable heterogeneity in the effects of the exposure on the mediator and the mediator on the outcome does not lead to bias in the estimates.

    “Nevertheless, a simulation study demonstrates that even considerable heterogeneity in these effects does not lead to bias in the estimates.”
    • Also stated in: § Abstract · hedged
      “although simulation analyses suggest that estimates may be robust to substantial random heterogeneity.”
    Quote in the paperSubstantial claim
  • Findingown data

    Using multiple-stage least squares, the direct effect estimate is 0.053 (standard error 0.037).

    “Using the multiple-stage least squares method, again adjusting for age, sex and centre in all the regression stages, the estimates of total, direct and indirect effect are 0.052 (0.033), 0.053 (0.037) and −0.001 (0.017), respectively.”
    Quote in the paperNot a substantial claim
  • Findingown data

    No substantial bias was found when there was a zero mean interaction between X and Z in the model for Y.

    “A similar finding of no substantial bias was found when there was a zero mean interaction between X and Z in the model for Y (Web Table A1, available as Supplementary data at IJE online)”
    Quote in the paperNot a substantial claim
  • Findingown data

    Using structural equation modelling, the direct effect estimate is 0.048 (standard error 0.034).

    “Estimates of total, direct and indirect effect are 0.052 (0.032), 0.048 (0.034) and 0.004 (0.013), respectively.”
    Quote in the paperSubstantial claim
  • Findingreasoning

    Using equation (2), the direct effect of BMI on uric acid not via CRP is 0.053 (standard error 0.035).

    “Using equation (2), the direct effect of BMI on uric acid not via CRP is 0.053 (standard error 0.035).”
    Quote in the paperSubstantial claim
  • May combine two claims. 16
  • The quote only partly states this claim. 1
  • Classified as probably not a substantial claim. 14

  • May combine two claims.

    Findingown data

    A causal effect of Z on X was found in 4.8% of datasets, which is no more than would be expected by chance alone.

    “A causal effect of Z on X was found in 4.8% of datasets, which is no more than would be expected by chance alone.”
    Quote in the paperNot a substantial claim

  • May combine two claims. The quote only partly states this claim.

    Findingown data

    In the simulation study, mean estimates of the direct effect of X on Y were approximately 1.00 and mean estimates of the indirect effect were approximately 0.98 across all levels of heterogeneity.

    “1 | 1 | 1.00 | 0.19 | 0.19 | 1.01 | 0.20 | 0.19 | 1.00 | 0.23 | 0.22”
    • Also stated in: § Results · plain
      “1 | 1 | 0.98 | 0.20 | 0.19 | 0.98 | 0.20 | 0.19 | 0.98 | 0.21 | 0.20”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Premisecitation

    Mendelian randomization uses genetic variants, assumed to be instrumental variables for a particular exposure, to estimate the causal effect of that exposure on an outcome.

    “Mendelian randomization uses genetic variants, assumed to be instrumental variables for a particular exposure, to estimate the causal effect of that exposure on an outcome.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [1] Davey Smith G Ebrahim S ‘Mendelian randomization’: can genetic epidemiology contribute to understanding environmental determinants of disease? Int J Epidemiol 2003; 32: 1– 22. 12689998 source

      Not checked in the free run.

  • Classified as probably not a substantial claim. May combine two claims.

    Interpretationcitation

    Assuming the path diagram is correctly specified, coefficients from a SEM can be viewed as representing causal effects, although the causal nature of the estimates is by prior assumption rather than empirically established by the data.

    “Assuming that the path diagram is correctly specified, coefficients from a SEM can be viewed as representing causal effects, although the causal nature of the estimates is by prior assumption rather than being empirically established by the data.25”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [25] Cox D Causality: some statistical aspects. J R Stat Soc A 1992; 155: 291– 301.

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • The authors state this as an assumption: The path diagram in the structural equation model is correctly specified.

  • May combine two claims.

    Premisenone given

    If multiple risk factors with associated IVs are measured in the same dataset, the causal effect of each risk factor on the outcome can be estimated, and each risk factor can be considered as the outcome to estimate causal effects between risk factors.

    “If multiple risk factors with associated IVs have been measured in the same dataset, then the causal effect of each of the risk factors on the outcome can be estimated. Additionally, each of the risk factors can be considered as the outcome in an IV analysis, and the causal effects of the risk factors on each other can be estimated.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Premisenone given

    Parameters in causal networks can be estimated using structural equation models.

    “Alternatively, parameters in this causal network and other more complex networks can be estimated using structural equation models (SEMs).”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationnone given

    The methods have potential for the investigation of more complex networks between multiple interrelated exposures and disease outcomes.

    “have potential for the investigation of more complex networks between multiple interrelated exposures and disease outcomes.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation

    The instrumental variable assumptions are: association with the exposure of interest; lack of association with any confounder of the exposure-outcome relationship, including unmeasured confounders; and lack of conditional association with the outcome given the exposure and all confounders.

    “These variants must satisfy the assumptions of an instrumental variable (IV): association with the exposure of interest; lack of association with any confounder of the exposure–outcome relationship (including those that are unmeasured); and lack of conditional association with the outcome given the exposure and all the confounders.2”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [2] Didelez V Sheehan N Mendelian randomization as an instrumental variable approach to causal inference. Stat Methods Med Res 2007; 16: 309– 30. 17715159 source

      Not checked in the free run.

  • May combine two claims.

    Interpretationreasoning

    Changes in the effects of the exposure and mediator over time and feedback between the exposure and mediator cannot be addressed by a conventional Mendelian randomization analysis.

    “As such, changes in the effects of the exposure and mediator over time and feedback between the exposure and mediator cannot be addressed by a conventional Mendelian randomization analysis.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [32] Burgess S Butterworth A Malarstig A Thompson S Use of Mendelian randomisation to assess potential benefit of clinical intervention. BMJ 2012; 345: e7325. 23131671 source

      Not checked in the free run.

  • May combine two claims.

    Interpretationcitation

    Causal effects estimated in a Mendelian randomization analysis represent long-term relationships, equivalent to a randomized trial where the intervention is made at conception.

    “As genetic subgroups of a population defined by an IV represent subpopulations with long-term average differences in the exposure of interest,31 the causal effects estimated in a Mendelian randomization analysis represent long-term relationships, equivalent to a randomized trial where the intervention is made at conception.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [31] Ebrahim S Davey Smith G Mendelian randomization: can genetic epidemiology help redress the failures of observational epidemiology? Hum Genet 2008; 123: 15– 33. 18038153 source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The natural direct effect is intrinsically a counterfactual quantity and cannot be observed from any experiment.

    “This is intrinsically a counterfactual quantity, and as such cannot be observed from any experiment.33”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [33] Robins J Richardson T Causality and psychopathology: finding the determinants of disorders and their cures. Alternative Graphical Causal Models and the Identification of Direct Effects. New York: Oxford University Press, 2011.

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Estimation of parameters and many goodness-of-fit tests in SEM rely on the assumption of multivariate normality of the variables.

    “Estimation of parameters and many goodness-of-fit tests rely on the assumption of multivariate normality of the variables.29”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [29] Muthén B Beyond SEM: General latent variable modeling. Behaviormetrika 2002; 29: 81– 117.

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • The authors state this as an assumption: The variables are multivariate normal for SEM estimation and goodness-of-fit tests.

  • May combine two claims.

    Study design and conductcitation

    With a sample size of 5000 and the simulated IV effects, the average F statistics are around 65, so the potential of bias from weak instruments is small.

    “With a sample size of 5000, this corresponds to average F statistics of around 65, so the potential of bias from weak instruments is small.38”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [38] Burgess S Thompson S; CRP CHD Genetics Collaboration. Avoiding bias from weak instruments in Mendelian randomization studies. Int J Epidemiol 2011; 40: 755– 64. 21414999 source

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Premisecitation

    The total effect of the exposure on the outcome is equal to the sum of the natural direct and indirect effects.

    “The total effect of the exposure on the outcome is equal to the sum of the natural direct and indirect effects.11,13”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [11] Pearl J Direct and indirect effects. Proceedings of the Seventeenth Conference on Uncertainty in Artificial Intelligence, Seattle, WA, 2–5 August 2001. San Francisco, CA: Morgan Kaufmann, 2001. source

      Not checked in the free run.

    • cites [13] Robins J Greenland S Identifiability and exchangeability for direct and indirect effects. Epidemiology 1992; 3: 143– 55. 1576220 source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductown data

    The Monte Carlo standard error of the mean estimates is around 0.005.

    “The Monte Carlo standard error of the mean estimates, representing the uncertainty due to the finite number of datasets, is around 0.005.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    The regression-based and SEM methods give similar estimates, which are consistent in the presence of unmeasured confounding, under the instrumental variable assumptions together with further assumptions on the linearity of effects without interaction terms and on the homogeneity of individual-level effects of the exposure on the mediator and the exposure and mediator on the outcome.

    “The regression-based and SEM methods discussed in this paper give similar estimates, which are consistent in the presence of unmeasured confounding, under the instrumental variable assumptions together with further assumptions on the linearity of effects without interaction terms and on the homogeneity of individual-level effects of the exposure on the mediator and the exposure and mediator on the”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: Genetic variants are available which satisfy the instrumental variable assumptions for the exposure–mediator and mediator–outcome relationships.
    • Rests on an assumption: The data-generating models considered in the simulation are the most likely to occur in applied examples.
    • The authors state this as an assumption: Effects are linear without interaction terms.

  • Classified as probably not a substantial claim.

    Study design and conductnone given

    In each regression, adjustment is made for age, sex and centre.

    “In each regression, adjustment is made for age, sex and centre.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    Increases in BMI cause increases in CRP levels, but the opposite is not true.

    “suggesting that increases in BMI cause increases in CRP levels, but the opposite is not true.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: Genetic variants are available which satisfy the instrumental variable assumptions for the exposure–mediator and mediator–outcome relationships.

  • May combine two claims.

    Interpretationown data

    Estimates of the direct and indirect effects using the methods presented in this paper are robust to quite substantial random heterogeneity in the causal and genetic effects, and to random (zero mean) individual-level interaction, in the range of simulation examples considered.

    “We conclude that estimates of the direct and indirect effects using the methods presented in this paper are robust to quite substantial random heterogeneity in the causal and genetic effects, and to random (zero mean) individual-level interaction, in the range of simulation examples considered.”
    • Also stated in: § Discussion · hedged
      “A simulation study suggests that random heterogeneity in the effects between the exposure, mediator and outcome does not lead to substantial bias in the estimators of the direct and indirect effects for the wide range of data-generating mechanisms considered; although there was some bias when variability in the individual-level effect parameters was correlated.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The data-generating models considered in the simulation are the most likely to occur in applied examples.

  • May combine two claims.

    Premisecitation

    Previous research has shown that genetic variants associated with BMI are associated with CRP levels and uric acid concentrations, but the reverse was not found for genetic variants which are plausible instrumental variables for CRP and uric acid.

    “Previous research has shown that genetic variants associated with BMI are associated with CRP levels30,39 and associated with uric acid concentrations,40 although in both cases the reverse was not found for genetic variants which are plausible IVs for CRP and for uric acid.”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites [30] Timpson N Nordestgaard B Harbord R C-reactive protein levels and body mass index: elucidating direction of causation through reciprocal Mendelian randomization. Int J Obes Relat Metab Disord 2011; 35: 300– 08. PMC4783860 20714329 source

      Not checked in the free run.

    • cites [39] Welsh P Polisecki E Robertson M Unraveling the directional link between adiposity and inflammation: a bidirectional Mendelian randomization approach. J Clin Endocrinol Metab 2010; 95: 93– 99. 19906786 PMC2805500 source

      Not checked in the free run.

    • cites [40] Palmer TM Nordestgaard BG Benn M Association of plasma uric acid with ischaemic heart disease and blood pressure: Mendelian randomisation analysis of two large cohorts. BMJ 2013; 347: f4262. 23869090 PMC3715134 source

      Not checked in the free run.

  • May combine two claims.

    Interpretationreasoning

    The greater efficiency of the SEM method corresponds to its stronger distributional assumption of multivariate normality, which is satisfied in this example.

    “This corresponds to the stronger distributional assumption of multivariate normality, which is satisfied in this example, made by the SEM method.”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: Individual-level effects of the exposure on the mediator and of the exposure and mediator on the outcome are homogeneous.

  • May combine two claims.

    Premisecitation

    Although not all causal models can be distinguished on the basis of observational data, models which have different conditional independence structures result in joint distributions for the variables which can be empirically compared.

    “Although not all causal models can be distinguished on the basis of observational data, models which have different conditional independence structures result in joint distributions for the variables which can be empirically compared.56,57”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [56] Pearl J An introduction to causal inference. Int J Biostat 2010; 6: 1– 60. PMC2836213 20305706

      Not checked in the free run.

    • cites [57] Rosa G Valente B de los Campos G Wu X Gianola D Silva M Inferring causal phenotype networks using structural equation models. Genet Sel Evol 2011; 43: 6. 21310061 PMC3056759

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S4 (based).

    Premisenone given

    In Mendelian randomization, the genetic variant is not permitted to have an effect on the outcome except via the exposure.

    “In Mendelian randomization, the genetic variant is not permitted to have an effect on the outcome except via the exposure.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Interpretationnone given

    A key difficulty in two-step epigenetic Mendelian randomization is finding genetic variants specifically associated with the phenotype and with the epigenetic marker if the two variables are closely biologically related.

    “A key difficulty here is finding genetic variants specifically associated with the phenotype and with the epigenetic marker if the two variables are closely biologically related.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationcitation

    When using genetic variants as instruments, the association between the instrumental variable and the exposure is often weak, and using the interaction between this variant and a baseline covariate as an instrumental variable for the mediator would typically result in an even weaker instrumental variable for the mediator, leading to substantial finite sample bias and imprecision, even if all theIV

    “When using genetic variants as instruments, the association between the IV and the exposure is often weak, and thus using the interaction between this variant and a baseline covariate as an IV for the mediator would typically result in an even weaker IV for the mediator, leading to substantial finite sample bias and imprecision, even if all the IV assumptions were met.50”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [50] Burgess S Thompson S Bias in causal estimates from Mendelian randomization studies with weak instruments. Stat Med 2011; 30: 1312– 23. 21432888 source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

    Premisenone given

    Although the causal network is generally high-dimensional and unknown, the direction of potential causal relationships between layers of data can often be deduced from external biological knowledge.

    “Although the causal network is generally high-dimensional and unknown, the direction of potential causal relationships between layers of data can often be deduced from external biological knowledge.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationcitation

    The use of instrumental variables provides a valuable addition to mediation analysis by relaxing the no unmeasured confounding assumption, but it does so at the cost of reintroducing stronger assumptions of linearity and no interaction, a trade-off which must carefully be evaluated in any given context.

    “The use of instrumental variables provides a valuable addition to mediation analysis by relaxing the no unmeasured confounding assumption, but it does so at the cost of reintroducing these stronger assumptions of linearity and no interaction, a trade-off which must carefully be evaluated in any given context.52”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [52] Ogburn E Commentary of ‘Mediation analysis without sequential ignorability: Using baseline covariates interacted with random assignment as instrumental variables’ by Dylan Small. J Stat Res 2012; 46: 105– 11. 25435642 PMC4244702

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S4 (based).

    Premisenone given

    Relationships between epigenetic markers, transcription factors and proteins can be affected by confounding and reverse causation in the same way as relationships between phenotypic exposures and outcomes.

    “Relationships between epigenetic markers, transcription factors and proteins can be affected by confounding and reverse causation in the same way as relationships between phenotypic exposures and outcomes.”
    Quote in the paperNot a substantial claim
28 of 28

  • The authors state this as an assumption.

    The causal effects of the exposure on the mediator and of the mediator on the outcome are homogeneous across individuals.

    “under the assumption of homogeneity of causal effects across individuals in the population”

  • The authors state this as an assumption.

    The path diagram in the structural equation model is correctly specified.

    “Assuming that the path diagram is correctly specified”

  • The authors state this as an assumption.

    The variables are multivariate normal for SEM estimation and goodness-of-fit tests.

    “Estimation of parameters and many goodness-of-fit tests rely on the assumption of multivariate normality of the variables.29”

  • The authors state this as an assumption.

    The causal effects between the exposure and mediator do not vary in direction over time.

    “all such estimates rely on the assumption that these effects do not vary in direction over time.”

  • The authors state this as an assumption.

    The causal effect of the exposure on the outcome is partially mediated by another risk factor.

    “We suppose that the causal effect of an exposure on an outcome is partially mediated by another risk factor.”

  • The authors state this as an assumption.

    Genetic variants are available which satisfy the instrumental variable assumptions for the exposure–mediator and mediator–outcome relationships.

    “Throughout, we have assumed that genetic variants are available which satisfy the IV assumptions for the exposure–mediator and mediator–outcome relationships.”

  • The paper relies on this without support.

    The data-generating models considered in the simulation are the most likely to occur in applied examples.

    “those considered in this paper were chosen as they were thought to be the most likely to occur in applied examples.”

  • The authors state this as an assumption.

    Effects are linear without interaction terms.

    “under the instrumental variable assumptions together with further assumptions on the linearity of effects without interaction terms and on the homogeneity of individual-level effects of the exposure on the mediator and the exposure and mediator on the outcome.”

  • The authors state this as an assumption.

    Individual-level effects of the exposure on the mediator and of the exposure and mediator on the outcome are homogeneous.

    “under the instrumental variable assumptions together with further assumptions on the linearity of effects without interaction terms and on the homogeneity of individual-level effects of the exposure on the mediator and the exposure and mediator on the outcome.”

  • The paper relies on this without support.

    The multivariate normality assumption of the SEM method is satisfied in the simulation example.

    “This corresponds to the stronger distributional assumption of multivariate normality, which is satisfied in this example, made by the SEM method.”
10 of 10

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • Combining a range of analyses with orthogonal validation is the most powerful way of addressing causal questions, within the triangulation of evidence framework.

    “Combining this with orthogonal validation – such as through the use of positive and negative controls ( Section 7) – is the most powerful way of addressing causal questions, within the triangulation of evidence framework 96, 97 .”

    From Guidelines for performing Mendelian randomization investigations (2020) · cites this paper as [96]

  • Mendelian randomization investigations are worthwhile in providing an alternative line of aetiological evidence even though the instrumental variable assumptions can never be proved beyond all doubt.

    “Mendelian randomization investigations are worthwhile in providing an alternative line of aetiological evidence even though the instrumental variable assumptions can never be proved beyond all doubt 96, 97 .”

    From Guidelines for performing Mendelian randomization investigations (2020) · cites this paper as [96]

Cites

Flagged first, then by citations

Research works printed but not linked: 6

  1. 2004, International Journal of Epidemiology, 3, 30
  2. 2009, Epidemiology (Cambridge, Mass.), 10, 18
  3. 2010, 25, 51, 10.1214/10-STS321 · DOI · Paper page
  4. 2012, J STAT RES, 46, 89
  5. 2012, J STAT RES, 46, 105
  6. 2010, 6, 1

Cited by

Sorted by citations

Related

Authors
Burgess, S., Daniel, R. M., Butterworth, A. S., Thompson, S. G., & the EPIC-InterAct Consortium
Affiliations
University of Cambridge (GB)
Corresponding author
Burgess, S.
ORCID
Venue
International Journal of Epidemiology
Publisher
Oxford University Press (OUP)
Published
2014-08-22
Volume
44
Issue
2
Pages
484-495
ISSN
  • 0300-5771 (ISSN-L)
  • 1464-3685
DOI
10.1093/ije/dyu176
PMC
PMC4469795
OpenAlex
W2111735528
Cited by
543
References
66
Citations by year
2026: 52 · 2025: 112 · 2024: 141 · 2023: 65 · 2022: 30 · 2021: 40 · 2020: 27 · 2019: 20 · 2018: 16 · 2017: 20 · 2016: 14 · 2015: 6
FWCI
14.10
Type
journal article
Language
en
Open access
Licence
cc-by
Locations
  • International Journal of Epidemiology (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • ORCA Online Research @Cardiff (Cardiff University) (acceptedVersion)
  • LSHTM Research Online (London School of Hygiene and Tropical Medicine) (acceptedVersion)
  • Apollo (University of Cambridge) (submittedVersion, cc-by)
Funders
  • Wellcome Trust (100114)
  • National Institute for Health and Care Research (NF-SI-0512-10165)
  • British Heart Foundation (SP/08/007/23628, RG/08/014/24067)
  • European Commission (LSHM_CT_2006_037197)
  • Medical Research Council (G1002283, MR/L003120/1)
Topics
  • Physical Sciences › Mathematics › Statistics and Probability › Advanced Causal Inference Techniques
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Genetics › Genetic Associations and Epidemiology
  • Physical Sciences › Mathematics › Statistics and Probability › Statistical Methods in Epidemiology
Keywords
mediation analysis, causal inference, genetic instruments, direct and indirect effects, unmeasured confounding, reverse causality, body mass index, C-reactive protein, uric acid
MeSH terms
Humans, Monte Carlo Method, Regression Analysis, Genetic Variation, Models, Statistical, Causality, Mendelian Randomization Analysis

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-10.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-10.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review