Pith. sign in

REVIEW 3 major objections 6 minor 49 references

This paper claims that an adaptive-lasso penalty on horizontal pleiotropy can make invalid-instrument selection consistent in two-sample Mendelian randomization with summary data, and that bootstrap-smoothing the post-selection estimator re

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-01 15:12 UTC pith:EF7UGIRO

load-bearing objection A useful and well-simulated extension of MR-Lasso with adaptive penalization, but the oracle-selection theorem rests on an unproven rate condition and the bootstrap inference is heuristic rather than exact. the 3 major comments →

arxiv 2607.18503 v1 pith:EF7UGIRO submitted 2026-07-20 stat.ME stat.AP

Adaptive Penalization and Bootstrap-Smoothed Inference for Two-Sample Mendelian Randomization with Summary Data

Muhammad Qasim , Kai Wang , Ishan S Bhatt This is my paper
classification stat.ME stat.AP MSC 62J0762F4062P10
keywords Mendelian randomizationhorizontal pleiotropyadaptive lassobootstrap smoothingpost-selection inferencesummary datainvalid instrumentsoracle property
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper tries to establish that two known weaknesses of the MR-Lasso approach—inconsistent identification of invalid genetic instruments and anti-conservative post-selection inference—can be fixed by two extensions. The first, MR-ALasso, replaces the uniform penalty on direct effects with adaptive weights derived from a median-ratio initial estimator; the authors prove that, under a majority-valid condition and a minimum direct-effect size, the invalid set is recovered with probability tending to one and the post-selection IVW estimator becomes asymptotically identical to the oracle IVW estimator. The second, MR-ALasso-B, averages post-selection IVW estimates over SNP-level bootstrap resamples and uses a delta-method variance estimator, showing in simulations that this restores near-nominal coverage and type-I error control. The paper matters because horizontal pleiotropy is the central threat to causal inference from genetic instruments, and existing penalized MR methods lack reliable uncertainty quantification after selection.

Core claim

Under Assumptions 1 and 2 (majority-valid instruments, nonzero direct effects bounded away from zero, and SNP-exposure effects bounded away from zero), the MR-ALasso estimator with adaptive weights ω_j = |α̃_j|^{-ν} identifies the true invalid set with probability tending to one, provided the tuning parameter satisfies λ_n = o(√n) and n^{(ν-1)/2}λ_n → ∞. The key mechanism is that the preliminary residuals α̃_j = β̂_Yj − θ̃ β̂_Xj, based on the median ratio estimator θ̃, are O_p(n^{-1/2}) for valid instruments but converge to a nonzero constant for invalid ones; hence the adaptive penalty diverges on valid instruments and remains bounded on invalid ones, producing oracle shrinkage. Consequentl

What carries the argument

The central object is the adaptive penalty weight vector ω_j = |α̃_j|^{-ν}, where α̃_j = β̂_Yj − θ̃ β̂_Xj are residual direct effects from the median ratio estimator θ̃. These weights diverge on truly valid instruments (since α̃_j = O_p(n^{-1/2})) and stay bounded on truly invalid ones (since α̃_j → α_j ≠ 0), so under the stated tuning-rate conditions the penalty dominates the loss on valid coordinates and becomes asymptotically negligible on invalid ones—giving the oracle selection behavior. The theoretical analysis profiles out the causal parameter θ, reducing the objective to a penalized quadratic in α with matrix Ĉ_n = S − Sβ̂_X(β̂_X^T S β̂_X)^{-1} β̂_X^T S, which is the object whose con

Load-bearing premise

The entire oracle-selection result depends on the median ratio estimator θ̃ being √n-consistent under the majority-valid condition in the two-sample summary-data setting, because the adaptive weights need the O_p(n^{ν/2}) divergence on valid instruments and boundedness on invalid ones.

What would settle it

A simulation or analytical counterexample where the simple median ratio estimate is biased due to InSIDE violations combined with invalid instruments that are weak relative to their pleiotropic effects—if selection consistency and oracle equivalence fail in such a setting, the proposition's premises cannot both hold.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Applied MR studies with sparse pleiotropy can use MR-ALasso to obtain a consistent set of valid instruments, and the post-selection IVW estimate inherits oracle behavior when the conditions of Proposition 3.1 hold.
  • MR-ALasso-B provides a practical way to compute interval estimates after selection, addressing the anti-conservative coverage that affects MR-Lasso and naive post-selection inference in finite samples.
  • The bootstrap-smoothing variance estimator directly quantifies instability of the selected instrument set under SNP-level perturbations, rather than conditioning on a single selected model.
  • The methods are implemented in the accompanying R package, making them directly usable for sensitivity analysis in routine two-sample MR pipelines.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The theory assumes the median ratio estimator is √n-consistent in the two-sample summary-data setting, but this is not proven; if invalid pleiotropic effects are small relative to instrument strength, the adaptive weights may not diverge fast enough, and the oracle selection result would fail.
  • The bootstrap smoothing is developed for a fixed number of instruments; in genome-wide analyses with many SNPs, the delta-method variance may behave differently, and the majority-vote aggregation rule has no formal error control (the paper acknowledges this).
  • A natural extension is to apply the same adaptive-weight mechanism to other robust MR estimators (e.g., MR-Egger or mode-based) to obtain consistent selection plus bootstrap-smoothed inference, though the majority-valid and bounded-strength assumptions would need re-examination.
  • A testable prediction is that MR-ALasso-B's coverage improvement is largest when the selection is unstable—i.e., when many SNPs lie near the valid–invalid boundary—and smallest when selection is deterministic; this could be verified by simulation with varying effect sizes.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. The paper proposes two regularized estimators for two-sample Mendelian randomization with summary data. MR-ALasso extends MR-Lasso by using adaptive penalty weights based on a median-ratio initial estimator, aiming to improve consistent identification of invalid instruments under the majority-valid condition. MR-ALasso-B applies SNP-level bootstrap smoothing to the post-selection IVW estimator and uses an Efron-style delta-method variance formula to construct confidence intervals. Theoretical results are claimed in Proposition 3.1 (selection consistency of the invalid set) and Corollary 3.1 (oracle equivalence of post-selection IVW), with additional results for the bootstrap variance estimator. The paper includes extensive simulations (Model-I individual-level, Model-II summary-data), a real-data bidirectional MR analysis across 22 traits, and an R package MRAlasso.

Significance. If the theoretical results hold, the paper makes a useful contribution to robust MR methodology: it addresses a well-known limitation of MR-Lasso (irrepresentable-condition failures and unstable selection) through adaptive weighting, and it proposes a practical, simulation-supported correction for post-selection inference. The simulation study is substantial (2000 replicates, balanced/directional pleiotropy, InSIDE violation, equal/stronger invalid instruments) and consistently shows RMSE/selection improvements of MR-ALasso over MR-Lasso and better coverage/type-I error of MR-ALasso-B. The authors are transparent about limitations, provide a public R package, and the real-data application illustrates the methods. However, the core selection-consistency theorem depends on an unproved √n-consistency rate for the median-ratio estimator in the summary-data setting, and the bootstrap-smoothed confidence interval is only empirically validated rather than theoretically justified for coverage.

major comments (3)
  1. [Prop. 3.1 / Suppl. B.2 (Eq. A3)] Proposition 3.1 assumes √n(θ̃−θ)=O_p(1) for the median-ratio estimator, but this rate is not established for the two-sample summary-data model. In Eq. (A3), √n(α̃−α)=√n(β̂Y−βY)−√n(θ̃−θ)β̂X−θ√n(β̂X−βX). The first and third terms are O_p(1) by Assumption 1, but the middle term is O_p(1) only if the assumed rate holds. The proof cites Windmeijer et al. (2019) only for consistency of the median, which does not imply √n-consistency; ratio estimators with estimated denominators can converge more slowly. Consequently, the divergence of the adaptive weights in (A5), the penalty dominance in (A7), and P(Â_n=A)→1 are not proven for the actual median-ratio-based MR-ALasso. Corollary 3.1 inherits this gap. Please add a lemma proving √n-consistency under Assumptions 1 and 2, or state the needed regularity conditions and provide a citation that covers the summary-data case.
  2. [Section 3.1 / Prop. 3.1] Proposition 3.1 conditions on a deterministic sequence λ_n satisfying (9), but Section 3.1 selects λ_n by a data-dependent heterogeneity stopping rule based on Cochran's Q and an RSE threshold. There is no result showing that the selected λ_n satisfies (9), nor simulation evidence about the realized λ_n sequence. Since the selection-consistency theorem applies to the estimator with a λ_n in (9), the theorem does not directly cover the implemented MR-ALasso algorithm. Please either adapt the theory to the stopping rule or add a result/justification linking the data-driven λ_n to (9).
  3. [Section 3.3.1 / Theorem 3.2, Eq. (13)] The variance estimator (13) converges (Prop. 3.2) to the delta-method variance of the bootstrap-smoothed functional conditional on the observed summary statistics. It measures sensitivity of the selected-valid-set estimator to SNP-level multinomial perturbations, not the sampling variability of (β̂Xj, β̂Yj) from the GWAS. Thus the Wald interval from (13) is not established to have nominal coverage for the causal parameter; Theorem 3.2 is a statement about the SNP-resampling functional, not about honest post-selection inference. The paper's simulation evidence is encouraging, and the Discussion appropriately cautions that this is 'not exact selective inference,' but given the abstract's coverage and type-I error claims, this distinction should be made explicit in the abstract and the theoretical status of MR-ALasso-B clarified.
minor comments (6)
  1. [Title page] Typo: 'Department of Satistics' should be 'Statistics'.
  2. [Eq. (7)-(8)] The notation S=diag(˜w_j) is used for inverse-variance weights; in Eq. (8) and Theorem 3.1, clarify that Ĉ_n is built from the same S, not a different weighting matrix.
  3. [Theorem 3.1] The phrase 'after minimizing over θ, ˆαAL satisfies' is imprecise; 'is a minimizer of' would be clearer.
  4. [Section 4] Capitalization of 'Model-I' and 'Model-II' is inconsistent in places; use one style throughout.
  5. [Section 5] There is a sentence fragment: 'spanning cardiovascular, anthropometric, immune, neurological/psychiatric, and social domains' repeats the earlier clause; consider editing.
  6. [Table 3] 'All invalid' is defined as A ⊆ Â_n, which does not exclude false positives; consider also reporting the exact recovery proportion P(Â_n=A) in the simulation tables.

Circularity Check

0 steps flagged

No material circularity: the central adaptive-lasso and bootstrap-smoothing derivations are conditional on explicit assumptions and externally benchmarked; only minor non-load-bearing self-citation.

full rationale

The paper's derivation chain is not circular. Proposition 3.1 is explicitly conditional: it assumes the median-ratio initial estimator is sqrt(n)-consistent, i.e. sqrt(n)(theta_tilde - theta)=O_p(1), and then Supplement B.2 derives the divergence of adaptive weights on valid instruments and boundedness on invalid instruments from that assumption together with Assumption 1. This is a standard two-stage estimation argument, not a fitting of the target result. Whether the median estimator actually achieves sqrt(n)-consistency in the two-sample summary-data setting is a genuine unproven premise — the paper cites Windmeijer et al. (2019) for consistency rather than proving the rate — but that is a proof-gap/correctness risk, not a circular reduction. Corollary 3.1 follows directly from selection consistency and does not re-import the oracle as an input. The bootstrap-smoothed variance in Theorem 3.2 and Proposition 3.2 is derived self-containedly from the multinomial nonparametric delta method in Supplements B.4-B.5; it targets a well-defined conditional covariance and is not a renamed empirical result. The paper's own Discussion limits the bootstrap procedure as 'a practical inferential correction rather than a full solution to post-selection inference' and states that full asymptotic theory for MR-ALasso-B is future work, which supports the non-circular interpretation. Simulation studies benchmark against IVW, MR-Egger, MR-Median, MR-Mode, MR-Lasso, and an oracle IVW, and the real-data claims are qualified. The self-citations (Qasim et al. 2025 for lasso-type IV selection; Qasim 2026 for the R package) are background/software references and are not load-bearing for the new theoretical claims. Thus the appropriate finding is no significant circularity, with a low score reflecting only minor self-citation.

Axiom & Free-Parameter Ledger

3 free parameters · 6 axioms · 0 invented entities

The central claims rest on majority-valid instruments, bounded instrument strength, minimum pleiotropy signal, and an unproved √n-consistency condition for the median-ratio initial estimator. The method also depends on hyperparameters ν, λ_n, and τ, none of which is given a formal data-driven optimality theory. No new physical or structural entities are introduced.

free parameters (3)
  • Adaptive penalty exponent ν
    Controls the adaptive weights ω_j = |αtilde_j|^{-ν} in Eq. (7); the paper says 'for a given value of ν' but no data-driven selection rule or sensitivity analysis is provided.
  • Tuning parameter λ_n
    Regularization strength for the adaptive Lasso. Proposition 3.1 requires λ_n satisfying Eq. (9), but Section 3.1's heterogeneity stopping rule (RSE > 1 plus a chi-squared cutoff) is not shown to produce such λ_n.
  • Aggregation threshold τ = 0.5
    Cutoff for the aggregated valid set in Algorithm 1 and Section 3.3.1. The paper acknowledges τ is a tuning parameter controlling stability versus misclassification.
axioms (6)
  • domain assumption GWAS summary estimates are normal, independent across SNPs after LD pruning (Assumption 1).
    Basis of the weighted-regression likelihood; the paper explicitly calls this a 'working approximation' because cross-SNP correlations are not exactly zero.
  • domain assumption Fewer than half of candidate instruments are invalid: s < J/2 (Assumption 2(i)).
    Majority-valid condition required for identification and used in Proposition 3.1 and the initial median estimator.
  • domain assumption Nonzero pleiotropic effects are bounded away from zero: min_{j∈A} |α_j| ≥ c_α (Assumption 2(ii)).
    Minimum signal strength needed for consistent selection; without it invalid instruments cannot be separated from valid ones asymptotically.
  • domain assumption SNP–exposure associations are bounded away from zero: min_j |β_{Xj}| ≥ c_X (Assumption 2(iii)).
    Excludes weak instruments and ensures the ratio-based initial estimator is well behaved.
  • ad hoc to paper The initial median-ratio estimator is √n-consistent: √n(θtilde − θ) = O_p(1).
    Stated in Proposition 3.1 and used in Supplement B.2 to make adaptive weights diverge on valid instruments, but not proven in the paper for the two-sample summary-data setting.
  • standard math n^{-1} Ĉ_n → C with the principal submatrix C_AA positive definite.
    Regularity condition needed for strict local convexity of the oracle loss; standard in adaptive-lasso proofs.

pith-pipeline@v1.3.0-alltime-deepseek · 36593 in / 15912 out tokens · 178283 ms · 2026-08-01T15:12:09.015768+00:00 · methodology

0 comments
read the original abstract

Two-sample Mendelian randomization (MR) uses genetic variants as instrumental variables to estimate causal effects from observational data using summary association statistics. However, horizontal pleiotropy can invalidate standard MR estimators and lead to biased causal inference. Pleiotropy-robust methods have been proposed to address this issue, including regularization-based approaches such as MR-Lasso. However, MR-Lasso may fail to identify invalid instruments consistently, and its post-selection inference can be unreliable. In this paper, we develop two lasso-type procedures for two-sample MR with summary-level data. The first, MR-ALasso, extends MR-Lasso by introducing adaptive penalty weights for pleiotropic effects in order to improve the identification of valid and invalid instruments. The second, MR-ALasso-B, combines adaptive lasso selection with bootstrap smoothing to improve post-selection inference. We establish theoretical results for MR-ALasso under the two-sample summary data framework, including invalid instrument identification consistency and oracle-type post-selection behavior. Simulation studies show that MR-ALasso generally improves upon MR-Lasso in estimation accuracy and invalid-instrument identification, whereas MR-ALasso-B substantially improves coverage and type-I error control relative to naive post-selection inference. A real-data application based on bidirectional analyses of multiple complex traits further illustrates the practical usefulness of the proposed methods. We provide an R package, MRAlasso, to facilitate implementation.

Figures

Figures reproduced from arXiv: 2607.18503 by Ishan S Bhatt, Kai Wang, Muhammad Qasim.

Figure 1
Figure 1. Figure 1: Directed acyclic graph for a single genetic variant in the MR framework. Here [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Distribution of estimated causal effects under different pleiotropy scenarios in [PITH_FULL_IMAGE:figures/full_fig_p020_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Bidirectional MR matrix summarizing results for 22 complex traits using multi [PITH_FULL_IMAGE:figures/full_fig_p028_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: UpSet plot summarizing overlap of significant associations [PITH_FULL_IMAGE:figures/full_fig_p029_4.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

49 extracted references · 1 canonical work pages

  1. [1]

    Journal of the American statistical Association , volume=

    Instrumental variables estimation with some invalid instruments and its application to Mendelian randomization , author=. Journal of the American statistical Association , volume=. 2016 , publisher=

    2016

  2. [2]

    Journal of the American Statistical Association , volume=

    On the use of the lasso for instrumental variables estimation with some invalid instruments , author=. Journal of the American Statistical Association , volume=. 2019 , publisher=

    2019

  3. [3]

    Statistical Methods in Medical Research , volume=

    LASSO-type instrumental variable selection methods with an application to Mendelian randomization , author=. Statistical Methods in Medical Research , volume=. 2025 , publisher=

    2025

  4. [4]

    Statistics in medicine , volume=

    A framework for the investigation of pleiotropy in two-sample summary data Mendelian randomization , author=. Statistics in medicine , volume=. 2017 , publisher=

    2017

  5. [5]

    Genetic epidemiology , volume=

    Mendelian randomization analysis with multiple genetic variants using summarized data , author=. Genetic epidemiology , volume=

  6. [6]

    Genetic epidemiology , volume=

    Consistent estimation in Mendelian randomization with some invalid instruments using a weighted median estimator , author=. Genetic epidemiology , volume=. 2016 , publisher=

    2016

  7. [7]

    International journal of epidemiology , volume=

    Robust inference in summary data Mendelian randomization via the zero modal pleiotropy assumption , author=. International journal of epidemiology , volume=. 2017 , publisher=

    2017

  8. [8]

    International journal of epidemiology , volume=

    ‘Mendelian randomization’: can genetic epidemiology contribute to understanding environmental determinants of disease? , author=. International journal of epidemiology , volume=. 2003 , publisher=

    2003

  9. [9]

    Statistics in medicine , volume=

    Mendelian randomization: using genes as instruments for making causal inferences in epidemiology , author=. Statistics in medicine , volume=. 2008 , publisher=

    2008

  10. [10]

    , title =

    Burgess, Stephen and Butterworth, Adam and Thompson, Simon G. , title =. Genetic Epidemiology , year =

  11. [11]

    International journal of epidemiology , volume=

    Use of allele scores as instrumental variables for Mendelian randomization , author=. International journal of epidemiology , volume=. 2013 , publisher=

    2013

  12. [12]

    , title =

    Burgess, Stephen and Thompson, Simon G. , title =. 2015 , address =

    2015

  13. [13]

    International Journal of Epidemiology , year =

    Bowden, Jack and Davey Smith, George and Burgess, Stephen , title =. International Journal of Epidemiology , year =

  14. [14]

    and Davey Smith, George and Bowden, Jack , title =

    Hartwig, Fernando P. and Davey Smith, George and Bowden, Jack , title =. International Journal of Epidemiology , year =

  15. [15]

    PloS one , volume=

    Robust methods in Mendelian randomization via penalization of heterogeneous causal estimates , author=. PloS one , volume=. 2019 , publisher=

    2019

  16. [16]

    Wellcome open research , volume=

    Guidelines for performing Mendelian randomization investigations: update for summer 2023 , author=. Wellcome open research , volume=

    2023

  17. [17]

    Nature reviews Methods primers , volume=

    Mendelian randomization , author=. Nature reviews Methods primers , volume=. 2022 , publisher=

    2022

  18. [18]

    Genetic epidemiology , volume=

    A comparison of robust Mendelian randomization methods using summary data , author=. Genetic epidemiology , volume=. 2020 , publisher=

    2020

  19. [19]

    Nature genetics , volume=

    Detection of widespread horizontal pleiotropy in causal relationships inferred from Mendelian randomization between complex traits and diseases , author=. Nature genetics , volume=. 2018 , publisher=

    2018

  20. [20]

    Statistics in medicine , volume=

    Pleiotropy robust methods for multivariable Mendelian randomization , author=. Statistics in medicine , volume=. 2021 , publisher=

    2021

  21. [21]

    Journal of Econometrics , volume=

    Select the valid and relevant moments: An information-based LASSO for GMM with many moments , author=. Journal of Econometrics , volume=. 2015 , publisher=

    2015

  22. [22]

    Journal of the American statistical association , volume=

    The adaptive lasso and its oracle properties , author=. Journal of the American statistical association , volume=. 2006 , publisher=

    2006

  23. [23]

    2011 , publisher=

    Statistics for high-dimensional data: methods, theory and applications , author=. 2011 , publisher=

    2011

  24. [24]

    Annals of Statistics , year=

    Least angle regression , author=. Annals of Statistics , year=

  25. [25]

    Econometrica , volume=

    Sparse models and methods for optimal instruments with an application to eminent domain , author=. Econometrica , volume=. 2012 , publisher=

    2012

  26. [26]

    Annals of Statistics , volume=

    Statistical inference in two-sample summary-data Mendelian randomization using robust adjusted profile score , author=. Annals of Statistics , volume=. 2020 , publisher=

    2020

  27. [27]

    The annals of applied statistics , volume=

    Selecting invalid instruments to improve Mendelian randomization with two-sample summary data , author=. The annals of applied statistics , volume=

  28. [28]

    Wang , title =

    Kai Wang and Grace Z. Wang , title =. 2025 , note =

    2025

  29. [29]

    International journal of epidemiology , volume=

    Improving the accuracy of two-sample summary-data Mendelian randomization: moving beyond the NOME assumption , author=. International journal of epidemiology , volume=. 2019 , publisher=

    2019

  30. [30]

    The American Journal of Human Genetics , volume=

    Constrained maximum likelihood-based Mendelian randomization robust to both correlated and uncorrelated pleiotropic effects , author=. The American Journal of Human Genetics , volume=. 2021 , publisher=

    2021

  31. [31]

    Nature communications , volume=

    A robust and efficient method for Mendelian randomization with hundreds of genetic variants , author=. Nature communications , volume=. 2020 , publisher=

    2020

  32. [32]

    International journal of epidemiology , volume=

    Improving the visualization, interpretation and analysis of two-sample summary data Mendelian randomization via the Radial plot and Radial regression , author=. International journal of epidemiology , volume=. 2018 , publisher=

    2018

  33. [33]

    Journal of the American Statistical Association , pages=

    Winner’s Curse Free Robust Mendelian Randomization with Summary Data , author=. Journal of the American Statistical Association , pages=. 2026 , publisher=

    2026

  34. [34]

    Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=

    Stability selection , author=. Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=. 2010 , publisher=

    2010

  35. [35]

    Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=

    Variable selection with error control: another look at stability selection , author=. Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=. 2013 , publisher=

    2013

  36. [36]

    Proceedings of the 25th international conference on Machine learning , pages=

    Bolasso: model consistent lasso estimation through the bootstrap , author=. Proceedings of the 25th international conference on Machine learning , pages=

  37. [37]

    Estimation and Accuracy after Model Selection

    Discussion of “Estimation and Accuracy after Model Selection” by Brad Efron , author=. Journal of the American Statistical Association , volume=

  38. [38]

    Journal of the American Statistical Association , volume=

    Estimation and accuracy after model selection , author=. Journal of the American Statistical Association , volume=. 2014 , publisher=

    2014

  39. [39]

    Statistica Sinica , pages=

    Observations on bagging , author=. Statistica Sinica , pages=. 2006 , publisher=

    2006

  40. [40]

    Proceedings of the National Academy of Sciences , volume=

    Mendelian randomization for causal inference accounting for pleiotropy and sample structure using genome-wide summary statistics , author=. Proceedings of the National Academy of Sciences , volume=. 2022 , publisher=

    2022

  41. [41]

    International journal of epidemiology , volume=

    MendelianRandomization: an R package for performing Mendelian randomization analyses using summarized data , author=. International journal of epidemiology , volume=. 2017 , publisher=

    2017

  42. [42]

    2026 , publisher =

    Qasim, Muhammad , title =. 2026 , publisher =. doi:10.5281/zenodo.20843253 , url =

    work page doi:10.5281/zenodo.20843253 2026

  43. [43]

    and Foley, Christopher N

    Broadbent, Jim R. and Foley, Christopher N. and Grant, Andrew J. and Mason, Amy M. and Staley, James R. and Burgess, Stephen , year =. Wellcome Open Research , doi =

  44. [44]

    New England journal of medicine , volume=

    Reduction in the incidence of type 2 diabetes with lifestyle intervention or metformin , author =. New England journal of medicine , volume=. 2002 , publisher=

    2002

  45. [45]

    The Journal of Clinical Hypertension , volume=

    Effect of weight loss on blood pressure changes in overweight patients: a systematic review and meta-analysis , author=. The Journal of Clinical Hypertension , volume=. 2023 , publisher=

    2023

  46. [46]

    Circulation , volume=

    Low-density lipoprotein cholesterol is predominantly associated with atherosclerotic cardiovascular disease events in patients with evidence of coronary atherosclerosis: the Western Denmark Heart Registry , author=. Circulation , volume=. 2023 , publisher=

    2023

  47. [47]

    The Lancet , volume=

    Randomised trial of cholesterol lowering in 4444 patients with coronary heart disease: the Scandinavian Simvastatin Survival Study (4S) , author=. The Lancet , volume=. 1994 , publisher=

    1994

  48. [48]

    The Lancet , volume=

    Blood pressure lowering for prevention of cardiovascular disease and death: a systematic review and meta-analysis , author=. The Lancet , volume=. 2016 , publisher=

    2016

  49. [49]

    PLoS genetics , volume=

    Getting to GRIPS with MR-Egger: Modelling directional pleiotropy independently of allele coding , author=. PLoS genetics , volume=. 2025 , publisher=

    2025