pith. sign in

arxiv: 2606.21708 · v1 · pith:SOUVVXW5new · submitted 2026-06-19 · 📊 stat.ME

Semi- and non-parametric approaches to individualized treatment regimes in the presence of causal mediation

Misha Dolmatov , Erica E.M. Moodie , David A. Stephens , Dipankar Bandyopadhyay This is my paper

Pith reviewed 2026-06-26 13:16 UTC · model grok-4.3

classification 📊 stat.ME
keywords individualized treatment regimescausal mediation analysisBayesian semiparametric estimationmultiple mediatorsnonparametric methodskidney allocationhepatitis C
0
0 comments X

The pith

Bayesian semiparametric estimators enable optimal individualized treatment rules targeting specific mediation pathways.

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

This paper introduces Bayesian semiparametric and nonparametric estimators for conditional mediation effects involving multiple mediators. These estimators are then applied to derive individualized treatment rules that optimize outcomes along particular causal pathways instead of the overall effect. A reader would care because this approach offers more flexibility in medical decision making, such as focusing on certain biological mechanisms while avoiding others, and is illustrated with an example of kidney allocation from hepatitis C positive donors.

Core claim

The authors develop novel Bayesian semiparametric and nonparametric estimators for conditional mediation effects in the presence of multiple mediators and demonstrate their use in estimating optimal individualized treatment regimes that target specific causal pathways.

What carries the argument

Bayesian semiparametric and nonparametric estimators for conditional mediation effects with multiple mediators, used to construct optimal ITRs targeting specific pathways.

If this is right

  • Optimal ITRs can be constructed to maximize effects through chosen mediators rather than all pathways.
  • The methods accommodate multiple mediators simultaneously.
  • Application to kidney allocation with hepatitis C positive donors illustrates practical utility.
  • Enhanced interpretability arises from focusing on specific causal mechanisms.

Where Pith is reading between the lines

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

  • These estimators could be applied to other areas like oncology or mental health where pathway-specific treatments matter.
  • Future work might compare these Bayesian methods to frequentist alternatives in terms of performance under model misspecification.
  • If identification assumptions hold in real data, the approach could lead to more nuanced personalized medicine strategies.

Load-bearing premise

Conditional mediation effects with multiple mediators are identifiable from the observed data under standard causal assumptions.

What would settle it

Observing that the estimated optimal ITRs do not match the true optimal rules in a simulation with known data generating process would falsify the utility of the estimators.

Figures

Figures reproduced from arXiv: 2606.21708 by David A. Stephens, Dipankar Bandyopadhyay, Erica E.M. Moodie, Misha Dolmatov.

Figure 1
Figure 1. Figure 1: Summary of estimated mediation effects in the testing subsample of [PITH_FULL_IMAGE:figures/full_fig_p035_1.png] view at source ↗
Figure 1
Figure 1. Figure 1: Mediation effect decomposition for 5 randomly sampled individuals in the test set [PITH_FULL_IMAGE:figures/full_fig_p060_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Mediation effect decomposition for 5 randomly sampled individuals in the test set [PITH_FULL_IMAGE:figures/full_fig_p063_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Mediation effect decomposition for 5 randomly sampled individuals in the test set [PITH_FULL_IMAGE:figures/full_fig_p065_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Mediation effect decomposition for 5 randomly sampled individuals in the test set [PITH_FULL_IMAGE:figures/full_fig_p068_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Distribution of blip parameter estimates in the linear setting across [PITH_FULL_IMAGE:figures/full_fig_p077_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Distribution of blip parameter estimates in the survival (AFT) setting across [PITH_FULL_IMAGE:figures/full_fig_p077_6.png] view at source ↗
read the original abstract

Individualized treatment rules (ITRs) map an individual patient's characteristics to their recommended treatment value. Typically, the optimal ITR is defined as the rule which maximizes a mean counterfactual outcome; the resulting ITR maximizes the effect of treatment along all causal pathways to the outcome, including indirect pathways through mediating variables. Although maximizing the total effect is often sufficient, explicitly incorporating causal mediation in an ITR analysis has several potential benefits such as enhanced interpretability, and additional flexibility in targeting specific causal pathways. For this purpose, we introduce novel Bayesian semiparametric and nonparametric estimators for conditional mediation effects in the presence of multiple mediators and show how they can be used to estimate optimal ITRs. We demonstrate the proposed methodology via an application to optimal kidney allocation with hepatitis C positive donors.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 3 minor

Summary. The paper introduces novel Bayesian semiparametric and nonparametric estimators for conditional mediation effects in the presence of multiple mediators. These estimators are used to construct optimal individualized treatment regimes (ITRs) that target specific causal pathways (rather than the total effect). Identification relies on standard no-unmeasured-confounding, positivity, and consistency assumptions; the approach is applied to kidney allocation from hepatitis C positive donors, with a sensitivity analysis for the confounding assumption.

Significance. If the estimators are consistent and the ITR optimization performs as claimed, the work offers a useful extension of mediation analysis to ITR settings, allowing pathway-specific targeting with potential benefits for interpretability in applications such as organ allocation. The explicit derivation under stated identification conditions and the sensitivity check are strengths.

major comments (1)
  1. [§3] §3 (estimator derivation): the Bayesian nonparametric component for multiple mediators requires a specific factorization or independence assumption across mediators conditional on covariates and treatment; without an explicit statement of this factorization or a robustness check when it is violated, it is unclear whether the conditional mediation effect estimator remains valid for the ITR objective in the kidney data.
minor comments (3)
  1. [§2 and §4] The notation distinguishing direct, indirect, and total effects across multiple mediators is introduced in §2 but reused without redefinition in the ITR section; a short table of symbols would improve readability.
  2. [Figure 2] Figure 2 (kidney allocation results) lacks error bars or credible intervals on the estimated value functions; adding these would allow direct visual assessment of uncertainty in the pathway-specific ITRs.
  3. [Introduction] The abstract states the estimators are 'novel' but the introduction does not explicitly contrast them with existing frequentist g-computation or IPW approaches for mediated ITRs; a one-sentence comparison would clarify the contribution.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their constructive comments on the manuscript. We address the single major comment below.

read point-by-point responses

  1. Referee: [§3] §3 (estimator derivation): the Bayesian nonparametric component for multiple mediators requires a specific factorization or independence assumption across mediators conditional on covariates and treatment; without an explicit statement of this factorization or a robustness check when it is violated, it is unclear whether the conditional mediation effect estimator remains valid for the ITR objective in the kidney data.

    Authors: In §3 the Bayesian nonparametric estimator models the joint conditional distribution of the multiple mediators given treatment and covariates via a Dirichlet process mixture that places no restriction on dependence structure. The factorization employed is the standard sequential one p(M1|A,X) p(M2|M1,A,X) … without any conditional-independence assumption across mediators. We will add an explicit statement of this factorization and the absence of an independence restriction in the revised manuscript. Because the model is fully nonparametric, the conditional mediation effect estimator remains valid for the subsequent ITR optimization under the identification assumptions already stated; the kidney-data application therefore inherits the same validity. A separate robustness check to factorization misspecification is not required given the nonparametric construction, but we can include a brief clarifying sentence if the referee prefers. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation is self-contained

full rationale

The manuscript states explicit identification assumptions (no unmeasured confounding for treatment-mediator, mediator-outcome, and treatment-outcome relations, plus positivity and consistency) and derives the Bayesian semiparametric and nonparametric estimators for conditional mediation effects under those conditions before constructing the ITR objective function. The kidney-allocation application includes a sensitivity analysis on the no-unmeasured-confounding assumption. No equation or step reduces by construction to a fitted parameter, self-citation chain, or renamed input; the central claims rest on independent derivation from the stated assumptions rather than tautological re-expression of the data.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review yields no identifiable free parameters, axioms, or invented entities; the central claim rests on unspecified identifiability conditions for mediation effects and correctness of the Bayesian models.

pith-pipeline@v0.9.1-grok · 5671 in / 966 out tokens · 25293 ms · 2026-06-26T13:16:31.258903+00:00 · methodology

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

99 extracted references

  1. [1]

    Murray and Ying Yuan and Peter F

    Thomas A. Murray and Ying Yuan and Peter F. Thall and Joan H. Elizondo and Wayne L. Hofstetter , journal =. A Utility-Based Design for Randomized Comparative Trials with Ordinal Outcomes and Prognostic Subgroups , volume =

  2. [2]

    , TITLE =

    Robins, James M. , TITLE =. Proceedings of the Second Seattle Symposium on Biostatistics , publisher =. 2004 , pages =

    2004

  3. [3]

    and Moodie, E

    Simoneau, G. and Moodie, E. E. M. and Nijjar, J. S. and Platt, R. W. and the Scottish Early Rheumatoid Arthritis Inception Cohort Investigators , title =. Journal of the American Statistical Association , volume =. 2020 , publisher =

    2020

  4. [4]

    Davison and David V

    Anthony C. Davison and David V. Hinkley. Bootstrap methods and their application. 1997

    1997

  5. [5]

    and Oakes, David

    Cox, David R. and Oakes, David. Analysis of Survival Data. 1984

    1984

  6. [6]

    and Limkemann, Ashley J

    Gupta, Gaurav and Kang, Le and Yu, Jonathan W. and Limkemann, Ashley J. and Garcia, Victoria and Bandyopadhyay, Dipankar and Kumar, Dhiren and Fattah, Hasan and Levy, Marlon and Cotterell, Adrian H. and Sharma, Amit and Bhati, Chandra and Reichman, Trevor and King, Anne L. and Sterling, Richard , title =. Clinical Transplantation , volume =

  7. [7]

    One-Step Generalized Estimating Equations With Large Cluster Sizes , volume =

    Lipsitz, Stuart and Fitzmaurice, Garrett and Sinha, Debajyoti and Hevelone, Nathanael and Hu, Jim and Nguyen, Louis , year =. One-Step Generalized Estimating Equations With Large Cluster Sizes , volume =

  8. [8]

    and Lizotte, Daniel J

    Laber, Eric B. and Lizotte, Daniel J. and Ferguson, Bradley , title =. Biometrics , volume =

  9. [9]

    and Lee, Sharon X

    McLachlan, Geoffrey J. and Lee, Sharon X. and Rathnayake, Suren I. Finite Mixture Models. Annual Review of Statistics and Its Application. 2019

    2019

  10. [10]

    Biometrika , volume =. 1986 , abstract =

    Liang, Kung-Yee and Zeger, Scott L. , title = ". Biometrika , volume =. 1986 , abstract = "

    1986

  11. [11]

    and van der Laan, Mark J and Robins, James M

    Gill, Richard D. and van der Laan, Mark J and Robins, James M. , editor =. Coarsening at Random: Characterizations, Conjectures, Counter-Examples , booktitle =. 1997 , publisher =

    1997

  12. [12]

    The Annals of Statistics , volume=

    Causal inference for complex longitudinal data: the continuous case , author=. The Annals of Statistics , volume=. 2001 , publisher=

    2001

  13. [13]

    Discussion of ``

    Rubin, Donald , journal=. Discussion of ``

  14. [14]

    Journal of the Royal Statistical Society Series A: Statistics in Society , volume =. 2016 , abstract =

    Yavuz, Idil and Cheng, Yu and Wahed, Abdus S. , title = ". Journal of the Royal Statistical Society Series A: Statistics in Society , volume =. 2016 , abstract = "

    2016

  15. [15]

    Biostatistics , volume =. 2018 , abstract =

    Chen, Ling-Wan and Yavuz, Idil and Cheng, Yu and Wahed, Abdus S. , title = ". Biostatistics , volume =. 2018 , abstract = "

    2018

  16. [16]

    Statistics in Medicine , volume =

    Morzywołek, Paweł and Steen, Johan and Van Biesen, Wim and Decruyenaere, Johan and Vansteelandt, Stijn , title =. Statistics in Medicine , volume =

  17. [17]

    2019 , author =

    Accelerated failure time models for the analysis of competing risks , journal =. 2019 , author =

    2019

  18. [18]

    Fine and Robert J

    Jason P. Fine and Robert J. Gray , journal =. A Proportional Hazards Model for the Subdistribution of a Competing Risk , volume =

  19. [19]

    Chakraborty, Bibhas and Moodie, E. E. M. , year =. Statistical Methods for Dynamic Treatment Regimes: Reinforcement Learning, Causal Inference, and Personalized Medicine , publisher=

  20. [20]

    and Davidian, Marie and Holloway, Shannon T

    Tsiatis, Anastasios A. and Davidian, Marie and Holloway, Shannon T. and Laber, Eric B. , year =

  21. [21]

    van der Vaart, A. W. , year=. Asymptotic Statistics , publisher=

  22. [22]

    Estimating Equations with Nuisance Parameters: theory and Applications , volume =

    Yuan, Ke-Hai and Jennrich, Robert , year =. Estimating Equations with Nuisance Parameters: theory and Applications , volume =

  23. [23]

    A General Approach to Causal Mediation Analysis , volume =

    Imai, Kosuke and Keele, Luke and Tingley, Dustin , year =. A General Approach to Causal Mediation Analysis , volume =

  24. [24]

    The Causal Mediation Formula--A Guide to the Assessment of Pathways and Mechanisms , volume =

    Pearl, Judea , year =. The Causal Mediation Formula--A Guide to the Assessment of Pathways and Mechanisms , volume =

  25. [25]

    The International Journal of Biostatistics , year =

    Optimal Individualized Treatments in Resource-Limited Settings , author =. The International Journal of Biostatistics , year =

  26. [26]

    R. L. Prentice and J. D. Kalbfleisch and A. V. Peterson and N. Flournoy and V. T. Farewell and N. E. Breslow , journal =. The Analysis of Failure Times in the Presence of Competing Risks , volume =

  27. [27]

    Bian, Zeyu and Moodie, E. E. M. and Shortreed, Susan and Bhatnagar, Sahir , year =. Variable Selection in Regression-Based Estimation of Dynamic Treatment Regimes , volume =

  28. [28]

    Pereira and Edgar L

    Brian J.G. Pereira and Edgar L. Milford and Robert L. Kirkman and Andrew S. Levey , title =. New England Journal of Medicine , volume =

  29. [29]

    , title =

    Murphy, Susan A. , title =. Journal of Machine Learning Research , pages =. 2005 , volume =

    2005

  30. [30]

    Wallace, M. P. and Moodie, E. E. M. , title =. Biometrics , year =

  31. [31]

    and Davidian, Marie , title =

    Lunceford, Jared K. and Davidian, Marie , title =. Statistics in Medicine , volume =

  32. [32]

    Biostatistics , volume =

    Simoneau, Gabrielle and Moodie, Erica E M and Platt, Robert W and Chakraborty, Bibhas , title = ". Biostatistics , volume =

  33. [33]

    Bayesian inference for optimal dynamic treatment regimes in practice , volume =

    Rodriguez Duque, Daniel and Moodie, Erica and Stephens, David , year =. Bayesian inference for optimal dynamic treatment regimes in practice , volume =

  34. [34]

    and Yuan, Ying and Thall, Peter , year =

    Murray, Thomas A. and Yuan, Ying and Thall, Peter , year =. A. Journal of the American Statistical Association , doi =

  35. [35]

    Cure Models in Survival Analysis

    Amico, Maïlis and Van Keilegom, Ingrid. Cure Models in Survival Analysis. Annual Review of Statistics and Its Application. 2018

    2018

  36. [36]

    Moodie, Erica E. M. and Stephens, David A and Alam, Shomoita and Zhang, Mei-Jie and Logan, Brent and Arora, Mukta and Spellman, Stephen and Krakow, Elizabeth F , journal=. A cure-rate model for. 2019 , publisher=

    2019

  37. [37]

    Lifetime Data Analysis , volume=

    Multi-state models: a review , author=. Lifetime Data Analysis , volume=. 1999 , publisher=

    1999

  38. [38]

    Health Economics , volume=

    A taxonomy of model structures for economic evaluation of health technologies , author=. Health Economics , volume=. 2006 , publisher=

    2006

  39. [39]

    Statistics in Medicine , volume=

    Reduced-rank proportional hazards regression and simulation-based prediction for multi-state models , author=. Statistics in Medicine , volume=. 2008 , publisher=

    2008

  40. [40]

    2018 , publisher=

    Multistate models for the analysis of life history data , author=. 2018 , publisher=

    2018

  41. [41]

    Rutter and Alan M

    Carolyn M. Rutter and Alan M. Zaslavsky and Eric J. Feuer , title =. Medical Decision Making , volume =

  42. [42]

    Advances in Neural Information Processing Systems , volume=

    Finite time analysis of stratified sampling for Monte Carlo , author=. Advances in Neural Information Processing Systems , volume=

  43. [43]

    48th annual conference on information sciences and systems (CISS) , pages=

    Best-arm identification algorithms for multi-armed bandits in the fixed confidence setting , author=. 48th annual conference on information sciences and systems (CISS) , pages=. 2014 , organization=

    2014

  44. [44]

    Conference on Learning Theory , pages=

    Optimal best arm identification with fixed confidence , author=. Conference on Learning Theory , pages=. 2016 , organization=

    2016

  45. [45]

    Randomized phase

    Estey, Elihu H and Thall, Peter F and Pierce, Sherry and Cortes, Jorge and Beran, Miloslav and Kantarjian, Hagop and Keating, Michael J and Andreeff, Michael and Freireich, Emil , journal=. Randomized phase. 1999 , publisher=

    1999

  46. [46]

    1957 , publisher=

    Dynamic programming , author=. 1957 , publisher=

    1957

  47. [47]

    Moodie, Erica E. M. and Richardson, Thomas S. , title =. Scandinavian Journal of Statistics , volume =

  48. [48]

    and Zhao, Yingqi , title =

    Chakraborty, Bibhas and Laber, Eric B. and Zhao, Yingqi , title =. Biometrics , volume =

  49. [49]

    Krakow, Elizabeth F. and Hemmer, Michael and Wang, Tao and Logan, Brent and Arora, Mukta and Spellman, Stephen and Couriel, Daniel and Alousi, Amin and Pidala, Joseph and Last, Michael and Lachance, Silvy and Moodie, Erica E. M. , title =. American Journal of Epidemiology , volume =

  50. [50]

    and Thall, Peter F

    Wahed, Abdus S. and Thall, Peter F. , title =. Journal of the Royal Statistical Society Series C: Applied Statistics , volume =

  51. [51]

    Murray and Brian P

    Thomas A. Murray and Brian P. Hobbs and Daniel J. Sargent and Bradley P. Carlin , title =. Bayesian Analysis , number =

  52. [52]

    and Lambert, Paul C

    Crowther, Michael J. and Lambert, Paul C. , title =. Statistics in Medicine , volume =

  53. [53]

    Epidemiology , volume=

    Interventional effects for mediation analysis with multiple mediators , author=. Epidemiology , volume=. 2017 , publisher=

    2017

  54. [54]

    Uncertainty in artificial intelligence , volume=

    Estimation of personalized effects associated with causal pathways , author=. Uncertainty in artificial intelligence , volume=

  55. [55]

    Epidemiology , volume=

    Effect decomposition in the presence of an exposure-induced mediator-outcome confounder , author=. Epidemiology , volume=. 2014 , publisher=

    2014

  56. [56]

    P. J. Bickel and F. Götze and W. R. van Zwet , journal =. RESAMPLING FEWER THAN n OBSERVATIONS: GAINS, LOSSES, AND REMEDIES FOR LOSSES , volume =

  57. [57]

    Rubin , journal =

    Donald B. Rubin , journal =. The

  58. [58]

    ESAIM: proceedings and surveys , volume=

    Introduction to vector quantization and its applications for numerics , author=. ESAIM: proceedings and surveys , volume=. 2015 , publisher=

    2015

  59. [59]

    Biometrika , volume=

    Decomposition, identification and multiply robust estimation of natural mediation effects with multiple mediators , author=. Biometrika , volume=. 2022 , publisher=

    2022

  60. [60]

    Psychometrika , volume=

    Robust inference for mediated effects in partially linear models , author=. Psychometrika , volume=. 2021 , publisher=

    2021

  61. [61]

    Journal of Causal Inference , volume=

    Nonparametric inference for interventional effects with multiple mediators , author=. Journal of Causal Inference , volume=. 2021 , publisher=

    2021

  62. [62]

    Journal of Causal Inference , volume=

    Heterogeneous interventional effects with multiple mediators: Semiparametric and nonparametric approaches , author=. Journal of Causal Inference , volume=. 2023 , publisher=

    2023

  63. [63]

    Electronic Journal of Statistics , number =

    Jean-Michel Loubes and Bruno Pelletier , title =. Electronic Journal of Statistics , number =

  64. [64]

    Handbook of

    Semiparametric doubly robust targeted double machine learning: a review , author=. Handbook of. 2024 , publisher=

    2024

  65. [65]

    Computational

    High-dimensional causal mediation analysis based on partial linear structural equation models , author=. Computational. 2022 , publisher=

    2022

  66. [66]

    Neurocomputing , volume=

    Double machine learning for partially linear mediation models with high-dimensional confounders , author=. Neurocomputing , volume=. 2025 , publisher=

    2025

  67. [67]

    , publisher =

    Hirano, Keisuke and Imbens, Guido W. , publisher =. The propensity score with continuous treatments , booktitle =

  68. [68]

    Proceedings of the 19th International Joint Conference on Artificial Intelligence , pages=

    Identifiability of path-specific effects , author=. Proceedings of the 19th International Joint Conference on Artificial Intelligence , pages=

  69. [69]

    arXiv preprint arXiv:2504.15834 , year=

    Causal machine learning for high-dimensional mediation analysis using interventional effects mapped to a target trial , author=. arXiv preprint arXiv:2504.15834 , year=

    arXiv

  70. [70]

    Handbook of Statistical Methods for Precision Medicine , pages=

    Selecting Optimal Subgroups for Treatment Using Many Covariates , author=. Handbook of Statistical Methods for Precision Medicine , pages=. 2024 , publisher=

    2024

  71. [71]

    Journal of Causal Inference , volume=

    Targeted data adaptive estimation of the causal dose--response curve , author=. Journal of Causal Inference , volume=. 2013 , publisher=

    2013

  72. [72]

    2018 , publisher=

    Double/debiased machine learning for treatment and structural parameters , author=. 2018 , publisher=

    2018

  73. [73]

    2000 , publisher=

    Foundations of quantization for probability distributions , author=. 2000 , publisher=

    2000

  74. [74]

    2002 , publisher=

    Principal component analysis , author=. 2002 , publisher=

    2002

  75. [75]

    Epidemiology , volume=

    Identifiability and exchangeability for direct and indirect effects , author=. Epidemiology , volume=. 1992 , publisher=

    1992

  76. [76]

    Epidemiologic methods , volume=

    Mediation analysis with multiple mediators , author=. Epidemiologic methods , volume=. 2014 , publisher=

    2014

  77. [77]

    , title =

    Tsybakov, Alexandre B. , title =

  78. [78]

    Annals of

    Minimax rates for heterogeneous causal effect estimation , author=. Annals of

  79. [79]

    Journal of the American Statistical Association , volume=

    Penalized estimating functions and variable selection in semiparametric regression models , author=. Journal of the American Statistical Association , volume=. 2008 , publisher=

    2008

  80. [80]

    Handbook of statistical methods for precision medicine , pages=

    Semiparametric doubly robust targeted double machine learning: a review , author=. Handbook of statistical methods for precision medicine , pages=. 2024 , publisher=

    2024

Showing first 80 references.