2021
DOI: 10.1101/2021.03.26.437168
|View full text |Cite
Preprint
|
Sign up to set email alerts
|

Bias correction for inverse variance weighting Mendelian randomization

Abstract: Inverse-variance weighted two-sample Mendelian Randomization (IVW-MR) is the most widely used approach that uses genome-wide association studies summary statistics to infer the existence and strength of the causal effect between an exposure and an outcome. Estimates from this approach can be subject to different biases due to: (i) the overlap between the exposure and outcome samples; (ii) the use of weak instruments and winner's curse. We developed a method that aims at tackling all these biases together. Assu… Show more

Help me understand this report
View published versions

Search citation statements

Order By: Relevance

Paper Sections

Select...
2
1
1
1

Citation Types

2
68
0

Year Published

2021
2021
2024
2024

Publication Types

Select...
6
4

Relationship

1
9

Authors

Journals

citations
Cited by 66 publications
(70 citation statements)
references
References 39 publications
2
68
0
Order By: Relevance
“…The positive causal e↵ect of smoking on MI, diabetes on asthma, the protective impact of higher birth weight on asthma, or higher education on smoking intensity, all of which were missed by standard MR could reflect the increased power of LHC-MR with its use of full-genome SNPs as opposed to genome-wide significant SNPs of classical MR approaches. Estimates from classical due to sample overlap is expected to be very small, and therefore not su cient to explain any discrepancy in the results [51] . Another possible reason for the discrepancy between our findings and those of standard MR methods is the presence of a significant heritable confounder found by LHC-MR with opposite e↵ect to the estimated causal e↵ect between the pair.…”
Section: Lhc Genetic Correlation Ldsc Genetic Correlationmentioning
confidence: 93%
“…The positive causal e↵ect of smoking on MI, diabetes on asthma, the protective impact of higher birth weight on asthma, or higher education on smoking intensity, all of which were missed by standard MR could reflect the increased power of LHC-MR with its use of full-genome SNPs as opposed to genome-wide significant SNPs of classical MR approaches. Estimates from classical due to sample overlap is expected to be very small, and therefore not su cient to explain any discrepancy in the results [51] . Another possible reason for the discrepancy between our findings and those of standard MR methods is the presence of a significant heritable confounder found by LHC-MR with opposite e↵ect to the estimated causal e↵ect between the pair.…”
Section: Lhc Genetic Correlation Ldsc Genetic Correlationmentioning
confidence: 93%
“…The two IV sets provided a partial confirmation of associations due to having different strengths; while IV set A was derived from a subsample of UK Biobank that did not overlap the sample used to estimate SNP-outcome effects, it is more likely to be underpowered, whereas IV set B maximizes power at the cost of introducing possible bias due to sample overlap. Recent simulations have indicated that bias due to sample overlap is minimal when using Mendelian randomization and is often negligible compared to other biases, such as winner's curse and weak-instrument bias [46,47]. We estimated potential bias due to sample overlap using the method proposed by Burgess et al [47], implemented in an online tool (https://sb452.shinyapps.io/overlap/).…”
Section: Genetic Associationsmentioning
confidence: 99%
“…However, considering potential interventions, effects of liability to educational attainment are much less useful than effects of specific education phenotypes. Fourth, the gene-exposure and gene-outcome estimates in the within-sibship GWAS two-sample MR analyses were from largely overlapping samples which could have potentially induced bias, although recent evidence suggests the degree of bias is likely to be modest 38 . Fifth, there is evidence that within-sibship models using PGS based on weighs from population GWAS could actually introduce bias 39 .…”
Section: Discussionmentioning
confidence: 99%