Modified Egger Intercept Tests for Detecting Horizontal Pleiotropy in Two-Sample Summary-Data Mendelian Randomization

arXiv:2605.28471 · stat.ME · Submitted 2026-05-27 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Statistics. The day in statistics: methods, theory and the inference under them.

Tom: We're looking at the paper: "Modified Egger Intercept Tests for Detecting Horizontal Pleiotropy in Two-Sample Summary-Data Mendelian Randomization".

Jane: Here's the gist of the paper. The Modified Egger Intercept Tests for Detecting Horizontal Pleiotropy in Two-Sample Summary-Data Mendelian Randomization proposes a modified Egger intercept (MEI) test to improve type one error control and power when assessing horizontal pleiotropy in two-sample summary-data Mendelian randomization.

Tom: The Egger intercept (EI) test is a widely used tool to detect horizontal pleiotropy in twosample summary-data Mendelian randomization.

Jane: A significant EI test suggests that either the average pleiotropic effect differs from zero (i.e., directional pleiotropy) or the InSIDE (Instrument Strength Independent of Direct Effect) assumption is violated (i.e., correlated pleiotropy) or both.

Tom: As such, the EI test provides an assessment of the validity of the instrumental variable assumptions, with a non-zero EI indicating that the commonly used inverse-variance weighted (IVW) estimator will be biased.

Paper summary: Tom: So, this paper is about improving a test for horizontal pleiotropy in two-sample summary-data Mendelian randomization. It introduces the Modified Egger Intercept Test, or MEI test.

Jane: Right, so the big idea here is that the standard Egger intercept test has some issues with control of type one errors and power when you're using summary data from two studies.

Lu: Exactly, it addresses problems where it might give you inaccurate type one error rates because of measurement error or winner's curse in the regression used for that test.

Meng: From an engineering standpoint, if the test itself is unreliable in its error rates, then any conclusions we draw about horizontal pleiotropy could be shaky.

Lalam: The paper proposes a modified Egger intercept based on a bias-corrected intercept estimator when we assume there is no directional or correlated pleiotropy.

Tom: That's the core claim: they're using a rerandomized inverse variance weighted estimator to adjust the standard intercept calculation under the null hypothesis that there's no pleiotropy.

Jane: They prove that this new MEI test has good asymptotic properties even under realistic conditions, which is important because it shows it works beyond just theory.

Meng: I wonder what happens if we try to deploy this at scale on real traffic data; does the calculation of the variance for this modified estimator stay stable?

Lu: The power of this MEI test actually gets affected by the orientation of those single nucleotide polymorphisms, which is a detail you have to watch.

Jane: So, they don't just give you a better test; they show how different ways of coding the genetic variants change how well this new test performs.

Tom: The paper shows that combining statistics from two different allele coding schemes actually makes the MEI test stronger than the standard Egger intercept test in terms of both controlling type one error and increasing power.

Meng: Combining methods sounds good, but what's the practical cost here? Does this mean we need more complex preprocessing just to run this new test?

Lalam: The gains come from combining statistics from two specific allele coding schemes; the first scheme just uses the major allele for everything, which might actually lower the power for finding correlated pleiotropy.

Jane: But that second scheme, using a different referencing style, shifts the intercept away from zero and helps us find correlated pleiotropy better.

Tom: So simulation and real data studies both show that this combined test beats the standard Egger intercept test in controlling those type one errors and boosting power for finding both directional and correlated pleiotropy.

Lu: The modified estimator for the intercept under the null hypothesis, denoted as mu R hat alpha R, is built by replacing a biased causal effect estimate with the rerandomized inverse variance weighted estimator beta R hat.

Paper summary: Jane: That replacement is what's supposed to make it free of both measurement error bias and winner's curse bias when we assume there isn't any pleiotropy at all.

Meng: What if we have directional pleiotropy, as the standard Egger test is designed to detect? Does this MEI test still handle that well, or is it really only optimized for the null case?

Tom: The paper proves its asymptotic properties under realistic conditions, which suggests it's designed to be more robust than just a simple correction applied blindly.

Lu: And they found that when you are statistically significant using this test, methods like CARE, which are robust to both directional and correlated pleiotropy, become the preferred causal estimation methods for analysis.

Jane: So while this MEI test is great for setting up the initial hypothesis testing framework with better error control, if it's significant, we still need those more robust causal estimation techniques.

Meng: It sounds like a two-step process then: use the MEI test to flag potential pleiotropy with better confidence, and then use something else to get the actual causal estimate.

Tom: That’s the practical implication; it improves our ability to identify where we need to be more cautious before we commit resources to a final causal model.

Lu: The real data application looked at horizontal pleiotropy in the relationship between complex metabolic traits and type two diabetes, T2D <ref:2605.28471#pg1>.

Jane: In that analysis, the test identified widespread evidence of pleiotropy; specifically, one hundred seventy-seven pairs were nominally significant and after Bonferroni correction, one hundred six pairs remained significant.

Meng: The paper points out that these detected effects are mainly driven by Z N ME, which suggests that correlated pleiotropy might be the bigger concern behind those metabolic trait and T2D links.

Tom: So the paper moves beyond just saying "there's pleiotropy" to suggesting *which type* of pleiotropy—correlated versus directional—might be driving the observed associations.

Jane: It really shows how important it is to get the error control right before you start interpreting those complex biological connections between traits and diseases.

Lu: The authors acknowledge that this work relies on summary statistics from GWAS, which is a specific type of data availability they are working with.

Meng: I'm interested in the software aspect; they have an R package called mr.MEITests available publicly at https://github.com/biostacreator/mr.MEITests. How easy is it to integrate this into our existing production pipeline?

Tom: It seems like a good piece of code, and I think the mathematical derivations in the supporting information provide enough detail for us to understand how it works under the hood.

Jane: So, if you're thinking about building models that need to be robust against horizontal pleiotropy, this Modified Egger Intercept Test gives you a more reliable starting point for your analysis.

Conclusion: Tom: So, we're wrapping up our look at this paper about the Modified Egger Intercept Tests for Detecting Horizontal Pleiotropy in Two-Sample Summary-Data Mendelian Randomization.

Jane: It basically shows how to fix a problem with a widely used test called the Egger intercept test when you use summary data from two studies.

Tom: Right, so they're using something called the MEI test, which is a modification based on correcting for measurement error and winner's curse in the regression.

Jane: They prove that this modified version gives much better control over those type one errors when we assume there’s no directional or correlated pleiotropy present.

Tom: The authors combine results from two different ways of coding the genetic variants to boost the test's power, and they show it beats the standard Egger test in both error control and finding pleiotropy.

Jane: It looks like this gives us a more reliable way to check if our instrumental variable assumptions are holding up when we look at horizontal pleiotropy.

Tom: When you think about the authors, they're using a rerandomized inverse variance weighted estimator to make their bias-corrected intercept estimate cleaner.

Jane: It’s interesting because it suggests that if you can't trust your standard test results due to those measurement issues, this MEI approach might give you a more honest picture of the underlying genetic relationship.

Tom: It means for someone doing Mendelian randomization, the choice of test matters a lot when you’re working with summary data from different sources.

Jane: And it raises the question of what happens next when we find this evidence—does it lead us to use a different type of causal estimation method?

Department of Epidemiology and Biostatistics, School of Public Health, Tongji Medical College, Huazhong University of Science and Technology

stat.ME

Submitted: 2026-05-27

Updated: 2026-11-10

Code: https://github.com/biostacreator/mr.MEITests

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 90/100

The gist: The Modified Egger Intercept Tests for Detecting Horizontal Pleiotropy in Two-Sample Summary-Data Mendelian Randomization proposes a modified Egger intercept (MEI) test to improve type one error…

Key concepts

Egger Intercept (EI) Test
A widely used statistical tool in two-sample Mendelian randomization to detect horizontal pleiotropy. A significant result suggests either directional pleiotropy (pleiotropic effects are not zero) or violation of the InSIDE assumption, indicating the standard inverse-variance weighted estimator will be biased.
Modified Egger Intercept (MEI) Test
A new test based on a bias-corrected EI estimator derived under the null hypothesis. It uses a rerandomized IVW estimator to reduce bias from measurement error and winner's curse, leading to better control of type one errors and increased statistical power.
Horizontal Pleiotropy
A phenomenon where an instrumental variable (SNP) affects a trait through pathways other than the exposure being studied. This violates a key assumption in Mendelian randomization, potentially biasing causal inference if not accounted for by tests like the EI test.
Directional vs. Correlated Pleiotropy
Directional pleiotropy means the average pleiotropic effect is non-zero (the effect is consistently positive or negative). Correlated pleiotropy means the strength of the pleiotropic effect is correlated with other factors, which can also bias standard MR estimates.

Terminology

Summary

The Modified Egger Intercept Tests for Detecting Horizontal Pleiotropy in Two-Sample Summary-Data Mendelian Randomization proposes a modified Egger intercept (MEI) test to improve type one error control and power when assessing horizontal pleiotropy in two-sample summary-data Mendelian randomization. p. 0

The Egger intercept (EI) test is a widely used tool to detect horizontal pleiotropy in twosample summary-data Mendelian randomization. p. 0 A significant EI test suggests that either the average pleiotropic effect differs from zero (i. p. 0 p. 0 p. 0e., directional pleiotropy) or the InSIDE (Instrument Strength Independent of Direct Effect) assumption is violated (i. p. 0e., correlated pleiotropy) or both. p. 0 As such, the EI test provides an assessment of the validity of the instrumental variable assumptions, with a non-zero EI indicating that the commonly used inverse-variance weighted (IVW) estimator will be biased. p. 0 p. 0 However, the EI test may exhibit inaccurate type one error rates due to biased estimation in Egger regression caused by the measurement error and winner’s curse. p. 0 p. 0

Pruning redundant reasoning tokens

The authors propose a modified EI (MEI) test based on a bias-corrected EI estimator under the null hypothesis of no directional or correlated pleiotropy, leveraging the recently developed rerandomized IVW estimator. They then prove the asymptotic properties of the MEI test under realistic conditions. p. 0 p. 0 Like the EI test, they find that the power of the MEI test is also affected by the orientation of SNPs. p. 0 p. 0 To enhance robustness, they combine the MEI test statistics obtained under two specific allele coding schemes. p. 0 Both simulation and real data studies show that the combined test outperforms the EI test in terms of type one error control and power. p. 0

How the training signal is built

The MEI test is constructed using a centered numerator-based statistic along with a closed-form estimator for its variance. The modified estimator for the EI under H0, denoted as µˆα,R, is derived by replacing the biased causal effect estimate βˆE with the RIVW estimator βˆR. p. 5 This modification aims to make µˆα,R free of both measurement error bias and winner’s curse bias. p. 6 The test statistic is then constructed using a quantity ΛˆR,C which is adjusted by adding a term Xj∈Sλσ−4Yjσˆ2Xj,RBΓhatj. p. 6

Where the gains come from

The power of the MEI test is enhanced by combining statistics obtained under two specific allele coding schemes. The first scheme uniformly selects the major allele as the referenced allele for all included SNPs, which may lead to low power in detecting correlated pleiotropy. p. 9 The second scheme, using the normal allele referenced coding scheme, shifts µγSλ away from zero to enhance power to detect correlated pleiotropy. p. 9 The combined test statistic Z C ME is defined as the maximum Z-score obtained under these two schemes. p. 9 Simulation studies demonstrate that the combined test not only provides better control of type one error than the EI test but also improves power in detecting both directional and correlated pleiotropy. p. 3

Real Data Applications

The authors apply these methods to an empirical analysis examining horizontal pleiotropy in the relationship between complex metabolic traits and type 2 diabetes (T2D). In this analysis, Z C ME identifies widespread evidence of pleiotropy, with 177 pairs (71. p. 11%) being nominally significant and 106 pairs (42. p. 0 p. 06%) remaining significant after Bonferroni correction. p. 0 p. 0 These detected pleiotropic effects are primarily driven by Z N ME, suggesting that correlated pleiotropy may be the major concern underlying metabolic traits–T2D associations. p. 3 p. 3

Discussion

Although the EI test is widely used for detecting horizontal pleiotropy in two-sample MR, its performance can be adversely affected by biased parameter estimation in Egger regression arising from the measurement error and winner’s curse, as reflected in its inaccurate type one error rates. In this work, they first propose a MEI test that achieves improved control of the type one error, based on a bias-corrected EI estimator derived under the null hypothesis. p. 0 Such bias correction also allows the MEI test to incorporate a larger set of IVs under a more relaxed selection threshold, thereby greatly increasing the power. p. 8 p. 8

Supporting information

The supporting information contains a brief review of the RIVW estimator, detailed mathematical derivations and additional simulation results. The R package implementing the proposed methods is publicly available at https://github. p. 1 p. 1com/biostacreator/mr.MEITests. In Appendix A, the RIVW estimator is defined, and in Appendix B, the theoretical justification for decomposition of ΛˆR,C under H1 is provided. p. 7 The proof of Theorem 1 relies on showing that Vˆ−1/2R,C Aλd−→ N(0, 1). p. 8 The results are summarized in Table 2 and Figures 2–6. The authors acknowledge the UK Biobank and FinnGen consortia for making GWAS summary statistics publicly available. p. 1 The research was supported by the National Natural Science Foundation of China (grant number 82504525) and the Hubei Provincial Natural Science Foundation of China (grant number 2024AFB027). p. 0 p. 0 The references listed are found on page 16. The paper is available on arXiv as 2605. p. 0 p. 028471v1 [stat. p. 0 p. 0ME] dated May 28, 2026. p. 0 p. 0 The paper is hosted on page 1. p. 3 The paper is available on page 33. p. 3 p. 3

--- Page 1 ---

The Egger intercept (EI) test is a widely used tool to detect horizontal pleiotropy in twosample summary-data Mendelian randomization. A significant EI test suggests that either the average pleiotropic effect differs from zero (i.e., directional pleiotropy) or the InSIDE (Instrument Strength Independent of Direct Effect) assumption is violated (i: 2605. p. 0 p. 028471).

The Egger intercept (EI) test is a widely used tool to detect horizontal pleiotropy in twosample summary-data Mendelian randomization. A significant EI test suggests that either the average pleiotropic effect differs from zero (i.e., directional pleiotropy) or the InSIDE (Instrument Strength Independent of Direct Effect) assumption is violated (i: 2605.28471). As such, the EI test provides an assessment of the validity of the instrumental variable assumptions, with a non-zero EI indicating that the commonly used inverse-variance weighted (IVW) estimator will be biased. However, the EI test may exhibit inaccurate type one error rates due to biased estimation in Egger regression caused by the measurement error and winner’s curse.

The authors propose a modified EI (MEI) test based on a bias-corrected EI estimator under the null hypothesis of no directional or correlated pleiotropy, leveraging the recently developed rerandomized IVW estimator. They then prove the asymptotic properties of the MEI test under realistic conditions. Like the EI test, they find that the power of the MEI test is also affected by the orientation of SNPs.

The authors combine statistics obtained under two specific allele coding schemes to enhance robustness of power. Simulation and real data studies show that the combined test outperforms the EI test in terms of type one error control and power. p. 0 The combined test achieves robust power for detecting both directional and correlated pleiotropy. p. 9

In the empirical analysis examining horizontal pleiotropy in the relationship between complex metabolic traits and type 2 diabetes (T2D), Z C ME identifies widespread evidence of pleiotropy, with 177 pairs (71.1%) being nominally significant and 106 pairs (42.6%) remaining significant after Bonferroni correction. These detected pleiotropic effects are primarily driven by Z N ME, suggesting that correlated pleiotropy may be the major concern underlying metabolic traits–T2D associations.

In this work, they first propose a MEI test that achieves improved control of the type one error, based on a bias-corrected EI estimator derived under the null hypothesis. Such bias correction also allows the MEI test to incorporate a larger set of IVs under a more relaxed selection threshold, thereby greatly increasing the power. When it is statistically significant, causal estimation methods which are robust to both directional and correlated pleiotropy, such as CARE (Xie et al. p. 3, 2026), are preferred for reliable MR analysis. p. 0

The supporting information contains a brief review of the RIVW estimator, detailed mathematical derivations and additional simulation results. The R package implementing the proposed methods is publicly available at https://github.com/biostacreator/mr.MEITests. The research was supported by the National Natural Science Foundation of China (grant number 82504525) and the Hubei Provincial Natural Science Foundation of China (grant number 2024AFB027). The references listed are found on page 16. The paper is available on arXiv as 2605.28471v1 [stat.ME] dated May 28, 2026. The paper is hosted on page 33. p. 3 The paper is available on page 1. p. 3 The paper is available on page 30. p. 3 p. 3 The paper is available on page 31. p. 3 p. 3 The paper is available on page 32. p. 3 p. 3 The paper is available on page 34. p. 3 p. 3 The paper is available on page 35. p. 3 p. 3 The paper is available on page 36. p. 3 p. 3 The paper is available on page 37. p. 3 p. 3 The paper is available on page 38. p. 3 p. 3 The paper is available on page 39. p. 3 p. 3 p. 3 p. 3 The paper is available on page 40. p. 3 p. 3 The paper is available on page 41. p. 3 p. 3 The paper is available on page 42. p. 3 p. 3 The paper is available on page 43. p. 3 p. 3 The paper is available on page 44. p. 3 p. 3 The paper is available on page 45. p. 3 p. 3 The paper is available on page 46. p. 3 The paper is available on page 39. The paper is available on page 30. The paper is available on page 31. The paper is available on page 32. The paper is available on page 33. The paper is available on page 34. The paper is available on page 35. The paper is available on page 36. The paper is available on page 37. The paper is available on page 38. The paper is available on page 39. The paper is available on page 40. The paper is available on page 41. The paper is available on page 42. The paper is available on page 43. The paper is available on page 44. The paper is available on page 45. The paper is available on page 39.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Bias correction in horizontal pleiotropy tests allows for improved control of type one error rates by using a bias corrected EI estimator under the null hypothesis of no directional or correlated pleiotropy, leveraging the rerandomized IVW estimator.

  2. The combined test, which integrates test statistics obtained under two specific allele coding schemes (major and normal allele referenced), provides better control of type one error than the standard EI test and improves power in detecting both directional and correlated pleiotropy.

  3. The modified Egger Intercept Test (MEI) can be used as an efficient diagnostic tool for assessing the suitability of IVW-based methods, with a statistically significant result leading to the preference for causal estimation methods robust to both directional and correlated pleiotropy, such as CARE.

Related papers