Multivariable MR can mitigate bias in two-sample MR using covariable-adjusted summary associations

Gilbody, Joseph; Borges, Maria Carolina; Smith, George Davey; Sanderson, Eleanor

doi:10.1101/2022.07.19.22277803

Cited by 16 publications

(14 citation statements)

References 33 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, for smoking or lung function-related phenotypes, adjusting for genotyping chip could additionally introduce collider/selection bias. Although such bias can theoretically be removed by using MVMR (40,41), prior research indicates that the use of MVMR in this context has little overall effect on the final MR estimates (40). Furthermore, adding additional covariates in an MVMR model greatly increases the risk of weak instrument bias and reduces power.…”

Section: Sensitivity Analyses For the Mr Estimation Approachesmentioning

confidence: 99%

WITHDRAWN: Assessing the Causal Effects of Environmental Tobacco Smoke Exposure: A meta-analytic Mendelian randomisation study

Woolf

Rajasundaram

Sallis

et al. 2023

Preprint

View full text Add to dashboard Cite

Objectives: Cigarette smoking is a major cause of global morbidity and premature mortality. There is evidence that exposure to environmental tobacco smoke (ETS; "second-hand smoking" or "passive smoking") contributes substantially to ill health. However, it is difficult to establish causality given well-described problems of confounding and selection bias. We applied Mendelian randomisation (MR) to investigate the causal effects of ETS exposure. Design: We employed six MR estimation approaches: the effects of each parent's smoking on their offspring, the effects of each parent's smoking on the other parent, and the effects of self-reported exposure to second-hand smoking both inside and outside of the home. Setting: Genome-wide association studies (GWASs) with sample sizes ranging from 397,732 to 909,629 individuals of predominantly European ancestry. Interventions: To mimic exposure to ETS in the first four approaches, we used an individual's exposure to their relative's genetic liability to smoke, conditional on the original individual's genetic liability to smoke. In the final two approaches we used instruments associated with self-reported exposure to ETS. Outcomes: Lung cancer, chronic obstructive pulmonary disease (COPD), stroke, cardiovascular disease, hypertension, and depression. Results: Our findings support a causal effect of genetically predicted ETS exposure on lung cancer and COPD (odds per hour exposed in a typical week = 1.41 [95% CI: 1.23 to 1.61] and 1.11 [95% CI: 1.07 to 1.16] respectively). We did not find evidence supporting an effect on other outcomes. Discussion: ETS may cause both lung cancer and COPD. Although our results do not support an effect of ETS on other outcomes, this might reflect low statistical power to detect smaller effect sizes. These results support public health measures to limit exposure to ETS.

show abstract

Section: Sensitivity Analyses For the Mr Estimation Approachesmentioning

confidence: 99%

WITHDRAWN: Assessing the Causal Effects of Environmental Tobacco Smoke Exposure: A meta-analytic Mendelian randomisation study

Woolf

Rajasundaram

Sallis

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…However, a recent study has shown that performing MR without adjusting for a factor that was conditioned on in one of the exposure or outcome GWAS can bias the resulting MR estimate, and adjusting for that factor can allow estimation of direct effects. (67) While age at first and last birth are conditioned on ever parous status since the GWAS was for these traits only being performed among parous women, the GWAS of breast cancer risk was not performed solely on parous women or adjusted for ever parous status, and so it is important to adjust for ever parous status in an MVMR model. Moreover, in the analysis of age at first birth on overall breast cancer risk, adjusting for ever parous status did modify the estimate to reveal an inverse effect, compared to the limited evidence identified in the univariable analyses.…”

Section: Discussionmentioning

confidence: 99%

“…While it may seem implausible to adjust for ever parous status in the MVMR analysis of age at first and last birth, since the age at first and last birth GWAS are conditioned on ever parous status by being performed only on parous women, while the breast cancer risk GWAS is not, it is important to adjust for ever parous status to reduce bias and allow estimation of direct effects. (67)…”

Section: Multivariable Analysismentioning

confidence: 99%

Evaluating the causal impact of reproductive factors on breast cancer risk: a multivariable mendelian randomization approach

Prince,

Howe,

Sanderson

et al. 2024

Preprint

Self Cite

View full text Add to dashboard Cite

Background: Observational evidence proposes a protective effect of having children and an early age at first birth on the development of breast cancer, however the causality of this association remains uncertain. In this study we assess whether these reproductive factors impact breast cancer risk independently of age at menarche, age at menopause, adiposity measures and other reproductive factors that have been identified as being causally related to or genetically correlated with the reproductive factors of interest. Methods: We used genetic data from UK Biobank (273,238 women) for reproductive factors, age at menarche and menopause, and adiposity measures, and the Breast Cancer Association Consortium for risk of overall, estrogen receptor (ER) positive and negative breast cancer as well as breast cancer subtypes. We applied univariable and multivariable Mendelian randomization (MR) to estimate direct effects of ever parous status, ages at first birth and last birth, and number of births on breast cancer risk. Results: We found limited evidence of an effect of age at first birth on overall or ER positive breast cancer risk in either the univariable or multivariable analyses. While the univariable analysis revealed an effect of later age at first birth decreasing ER negative breast cancer risk (Odds ratio (OR): 0.76, 95% confidence interval:0.61-0.95 per standard deviation (SD) increase in age at first birth), this effect attenuated with separate adjustment for age at menarche and menopause (e.g., OR 0.83, 0.62-1.06 per SD increase in age at first birth, adjusted for age at menarche). In addition, we found evidence for an effect of later age at first birth on decreased human epidermal growth factor receptor 2 enriched breast cancer risk but only with adjustment for number of births (OR 0.28 (0.11-0.57) per SD increase in age at first birth). We found little evidence for direct effects of ever parous status, age at last birth or number of births on breast cancer risk, however, analyses of ever parous status and age at last birth were limited by weak instruments in the multivariable analysis. Conclusions: This study found minimal evidence of a protective effect of earlier age at first birth on breast cancer risk, while identifying some evidence for an adverse effect on ER negative breast cancer risk. However, multivariable MR of ever parous status and age at last birth is limited by weak instruments which might be improved in future studies with larger sample sizes and when additional genetic variants related to reproductive factors are identified.

show abstract

“…We also showed that adjustment for BMI in GWAS may lead to an unexpected and misleading result in MR analysis (Supplementary Note 1), if BMI (i.e. heritable covariate) also has a role in the studied relationship [53] [54].…”

Section: Discussionmentioning

confidence: 99%

The mediating role of mammographic density in the protective effect of early-life adiposity on breast cancer risk: a multivariable Mendelian randomization study

Vabistsevits,

Smith,

Richardson

et al. 2023

Preprint

Self Cite

View full text Add to dashboard Cite

Observational studies suggest that mammographic density (MD) may have a role in the unexplained protective effect of childhood adiposity on breast cancer risk. Here, we investigated a complex and interlinked relationship between puberty onset, adiposity, MD, and their effects on breast cancer using Mendelian randomization (MR).We estimated the effects of childhood and adulthood adiposity, and age at menarche on MD phenotypes (dense area (DA), non-dense area (NDA), percent density (PD)) using MR and multivariable MR (MVMR), allowing us to disentangle their total and direct effects. Next, we examined the effect of MD on breast cancer risk, including risk of molecular subtypes, and accounting for genetic pleiotropy. Finally, we used MVMR to evaluate whether the protective effect of childhood adiposity on breast cancer was mediated by MD.Childhood adiposity had a strong inverse effect on mammographic DA, while adulthood adiposity increased NDA. Later menarche had an effect of increasing DA and PD, but when accounting for childhood adiposity, this effect attenuated to the null. DA and PD had a risk-increasing effect on breast cancer across all subtypes. The MD single-nucleotide polymorphism (SNP) estimates were extremely heterogeneous, and examination of the SNPs suggested different mechanisms may be linking MD and breast cancer. Finally, MR mediation analysis estimated that 56% (95% CIs [32% - 79%]) of the childhood adiposity effect on breast cancer risk was mediated via DA.In this work, we sought to disentangle the relationship between factors affecting MD and breast cancer. We showed that higher childhood adiposity decreases mammographic DA, which subsequently leads to reduced breast cancer risk. Understanding this mechanism is of great importance for identifying potential targets of intervention, since advocating weight gain in childhood would not be recommended.

show abstract

Multivariable MR can mitigate bias in two-sample MR using covariable-adjusted summary associations

Cited by 16 publications

References 33 publications

WITHDRAWN: Assessing the Causal Effects of Environmental Tobacco Smoke Exposure: A meta-analytic Mendelian randomisation study

WITHDRAWN: Assessing the Causal Effects of Environmental Tobacco Smoke Exposure: A meta-analytic Mendelian randomisation study

Evaluating the causal impact of reproductive factors on breast cancer risk: a multivariable mendelian randomization approach

The mediating role of mammographic density in the protective effect of early-life adiposity on breast cancer risk: a multivariable Mendelian randomization study

Contact Info

Product

Resources

About