pith. sign in

arxiv: 2601.19836 · v1 · submitted 2026-01-27 · 📊 stat.ME · stat.AP

Personalized Treatment Hierarchies in Bayesian Network Meta-Analysis

Augustine Wigle , Erica E. M. Moodie This is my paper

Pith reviewed 2026-05-16 10:43 UTC · model grok-4.3

classification 📊 stat.ME stat.AP
keywords network meta-analysistreatment hierarchiestreatment-covariate interactionsBayesian methodspersonalized rankingsmajor depressive disorder
0
0 comments X

The pith

Treatment hierarchies from network meta-analysis with interactions must be computed for a specific patient covariate profile instead of overall.

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

The paper shows that Bayesian network meta-analysis models incorporating treatment-covariate interactions require a change in how rankings are produced. Standard methods average effects across all patients, but with interactions the relative benefits shift depending on the value of a covariate such as age or baseline severity. The authors describe how to extract a hierarchy conditional on one chosen profile from the fitted model. They illustrate the difference using an existing network of trials for major depressive disorder treatments. This matters because an overall ranking can point clinicians toward a treatment that performs differently for the actual patients they see.

Core claim

In a Bayesian NMA that estimates treatment-covariate interactions, a treatment hierarchy is obtained by fixing the covariate at a clinically chosen value and then ranking the treatments according to the resulting posterior distributions of the relative effects. This replaces the conventional approach of ranking on the basis of marginal effects averaged over the covariate distribution.

What carries the argument

Treatment-covariate interactions (TCIs) inside a Bayesian network meta-analysis model, which allow the log-odds or mean differences between treatments to change linearly or otherwise with the covariate and thereby produce profile-specific rankings.

If this is right

  • A single network model can generate multiple hierarchies, one for each relevant patient profile, without requiring new data.
  • Overall rankings that ignore the covariate can mis-rank treatments for subgroups even when the model itself is correctly specified.
  • The method applies directly to any existing Bayesian NMA software that already supports TCIs.
  • In the depressive-disorder example the profile-specific orderings differed from the overall ordering for several treatments.

Where Pith is reading between the lines

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

  • Guidelines that currently publish one universal treatment ladder could instead publish a small set of ladders indexed by common covariate values.
  • Decision models that use NMA output as input would need to sample the covariate distribution rather than use a single set of probabilities.
  • The same logic extends to other synthesis methods that already contain interaction terms, such as individual-patient-data meta-analysis.

Load-bearing premise

The fitted interaction model correctly describes how treatment effects vary with the chosen covariate and the selected profile represents a clinically meaningful patient group.

What would settle it

Recalculate the hierarchies for two different covariate values within the observed range and check whether the ordering of treatments remains identical to the marginal ranking; if the order changes materially for realistic profiles, the claim holds.

Figures

Figures reproduced from arXiv: 2601.19836 by Augustine Wigle, Erica E. M. Moodie.

Figure 1
Figure 1. Figure 1: Posterior distributions of the relative effects of each treatment for Patients A and B. References 1. Salanti G, Nikolakopoulou A, Efthimiou O, Mavridis D, Egger M, and White IR. Introducing the Treatment Hierarchy Question in Network Meta-Analysis. American Journal of Epidemiology 2022;191:930–8. 2. Salanti G, Ades AE, and Ioannidis JPA. Graphical Methods and Numerical Summaries for Presenting Results fro… view at source ↗
read the original abstract

Network Meta-Analysis (NMA) is an increasingly popular evidence synthesis tool that can provide a ranking of competing treatments, also known as a treatment hierarchy. Treatment-Covariate Interactions (TCIs) can be included in NMA models to allow relative treatment effects to vary with covariate values. We show that in an NMA model that includes TCIs, treatment hierarchies should be created with a particular covariate profile in mind. We outline the typical approach for creating a treatment hierarchy in standard Bayesian NMA and show how a treatment hierarchy for a particular covariate profile can be created from an NMA model that estimates TCIs. We demonstrate our methods using a real network of studies for treatments of major depressive disorder.

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

2 major / 3 minor

Summary. The manuscript claims that in Bayesian network meta-analysis (NMA) models that include treatment-covariate interactions (TCIs), treatment hierarchies must be constructed conditional on a specific covariate profile rather than marginally. It outlines the standard Bayesian approach to hierarchies and extends it to profile-specific rank probabilities, demonstrating the method on a real network of studies for major depressive disorder (MDD) treatments.

Significance. If the central claim holds, the work is significant for evidence synthesis because it shows how to obtain clinically relevant, personalized treatment rankings when effect modifiers are present. This extends standard Bayesian NMA machinery in a practical way and could improve decision-making in heterogeneous populations without requiring new model parameters.

major comments (2)
  1. [§4 (MDD demonstration)] §4 (MDD demonstration): the paper describes the algebraic construction of profile-specific hierarchies but reports neither the marginal (covariate-averaged) hierarchy from the same posterior nor any numerical comparison (rank changes, SUCRA shifts, or P(best) differences) for the depressive-disorder network. This contrast is load-bearing for the claim that standard hierarchies are misleading once TCIs are included.
  2. [Methods section] Methods section: the outline of how profile-specific rank probabilities are obtained from the TCI model is given at a conceptual level but supplies no explicit equations, conditioning steps, or derivation showing how the posterior is evaluated at a fixed covariate value. Without these details the procedure cannot be verified or reproduced.
minor comments (3)
  1. The abstract would benefit from a one-sentence recap of the standard marginal hierarchy method with a key reference before describing the profile-specific extension.
  2. [Results] In the MDD results, the chosen covariate profile should be justified on clinical grounds and accompanied by a brief sensitivity check for nearby profiles.
  3. Ensure all acronyms (TCI, SUCRA, NMA) are defined at first use and that figure legends explicitly state the covariate value(s) used for each hierarchy.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments, which help clarify the presentation and strengthen the empirical demonstration of our approach. We address each major comment below.

read point-by-point responses

  1. Referee: [§4 (MDD demonstration)] the paper describes the algebraic construction of profile-specific hierarchies but reports neither the marginal (covariate-averaged) hierarchy from the same posterior nor any numerical comparison (rank changes, SUCRA shifts, or P(best) differences) for the depressive-disorder network. This contrast is load-bearing for the claim that standard hierarchies are misleading once TCIs are included.

    Authors: We agree that a direct comparison to the marginal hierarchy would make the practical consequences of conditioning on covariate profiles more compelling. In the revised manuscript we will compute and report the marginal (covariate-averaged) treatment hierarchy from the same posterior draws, together with numerical contrasts (rank changes, SUCRA differences, and shifts in P(best)) for the MDD network. These additions will be placed in §4 and will be accompanied by a brief discussion of when and why the conditional and marginal rankings diverge. revision: yes

  2. Referee: [Methods section] the outline of how profile-specific rank probabilities are obtained from the TCI model is given at a conceptual level but supplies no explicit equations, conditioning steps, or derivation showing how the posterior is evaluated at a fixed covariate value. Without these details the procedure cannot be verified or reproduced.

    Authors: We accept that the current description is too high-level for full reproducibility. In the revised Methods section we will insert the explicit derivation: (i) the linear predictor for the conditional treatment effect given covariate value x, (ii) the Monte Carlo procedure that evaluates the posterior at a fixed x to obtain the vector of relative effects, and (iii) the subsequent calculation of rank probabilities and SUCRA values from those conditional draws. The added equations will be numbered and cross-referenced to the existing notation. revision: yes

Circularity Check

0 steps flagged

No circularity; algebraic extension of standard Bayesian NMA

full rationale

The paper describes the standard Bayesian NMA treatment hierarchy construction and then algebraically extends it to profile-specific hierarchies by conditioning the posterior on a chosen covariate value within an existing TCI model. This step uses the fitted parameters directly without re-deriving them as new predictions, without self-definitional loops, and without load-bearing self-citations or imported uniqueness theorems. The demonstration on the MDD network applies the same posterior without reducing any reported quantity to a fitted input by construction. The derivation therefore remains self-contained against external Bayesian NMA benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available; no explicit free parameters, axioms, or invented entities are identifiable from the provided text.

pith-pipeline@v0.9.0 · 5409 in / 902 out tokens · 41113 ms · 2026-05-16T10:43:22.770890+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

19 extracted references · 19 canonical work pages

  1. [1]

    Introducing the Treatment Hierarchy Question in Network Meta-Analysis

    Salanti G, Nikolakopoulou A, Efthimiou O, Mavridis D, Egger M, and White IR. Introducing the Treatment Hierarchy Question in Network Meta-Analysis. American Journal of Epidemiology 2022;191:930–8

    work page 2022

  2. [2]

    Graphical Methods and Numerical Summaries for Presenting Results from Multiple-Treatment Meta-Analysis: An Overview and Tutorial

    Salanti G, Ades AE, and Ioannidis JPA. Graphical Methods and Numerical Summaries for Presenting Results from Multiple-Treatment Meta-Analysis: An Overview and Tutorial. Journal of Clinical Epidemiology 2011;64:163–71

    work page 2011

  3. [3]

    BMC Medical Research Methodology 2022;22:47

    PapakonstantinouT,SalantiG,MavridisD,RückerG,SchwarzerG,andNikolakopoulouA.AnsweringComplexHierarchy Questions in Network Meta-Analysis. BMC Medical Research Methodology 2022;22:47

    work page 2022

  4. [4]

    Combining Individual Patient Data and Aggregate Data in Mixed Treatment Comparison Meta-Analysis: Individual Patient Data May Be Beneficial If Only for a Subset of Trials

    Donegan S, Williamson P, D’Alessandro U, Garner P, and Smith CT. Combining Individual Patient Data and Aggregate Data in Mixed Treatment Comparison Meta-Analysis: Individual Patient Data May Be Beneficial If Only for a Subset of Trials. Statistics in Medicine 2013;32:914–30

    work page 2013

  5. [5]

    A Framework for Identifying Treatment-Covariate Interactions in Individual Participant Data Network Meta-Analysis

    Freeman SC, Fisher D, Tierney JF, and Carpenter JR. A Framework for Identifying Treatment-Covariate Interactions in Individual Participant Data Network Meta-Analysis. Research Synthesis Methods 2018;9:393–407. 6REFERENCES

    work page 2018

  6. [6]

    Using Individual Participant Data to Improve Network Meta-Analysis Projects

    Riley RD, Dias S, Donegan S, et al. Using Individual Participant Data to Improve Network Meta-Analysis Projects. BMJ Evidence-Based Medicine 2023;28:197–203

    work page 2023

  7. [7]

    Statistics in Medicine 2020;39:2115–37

    RileyRD,DebrayTP,FisherD,etal.IndividualParticipantDataMeta-AnalysistoExamineInteractionsbetweenTreatment Effect and Participant-Level Covariates: Statistical Recommendations for Conduct and Planning. Statistics in Medicine 2020;39:2115–37

    work page 2020

  8. [8]

    Twenty Years of Network Meta-Analysis: Continuing Controversies and Recent Developments

    Ades AE, Welton NJ, Dias S, Phillippo DM, and Caldwell DM. Twenty Years of Network Meta-Analysis: Continuing Controversies and Recent Developments. Research Synthesis Methods 2024;15

    work page 2024

  9. [9]

    Ranking Treatments in Frequentist Network Meta-Analysis Works without Resampling Methods

    Rücker G and Schwarzer G. Ranking Treatments in Frequentist Network Meta-Analysis Works without Resampling Methods. BMC Medical Research Methodology 2015;15:58

    work page 2015

  10. [10]

    BMC Medical Research Methodology 2021;21:213

    RosenbergerKJ,DuanR,ChenY,andLinL.PredictiveP-scoreforTreatmentRankinginBayesianNetworkMeta-Analysis. BMC Medical Research Methodology 2021;21:213

    work page 2021

  11. [11]

    Multilevel Network Meta-Regression for Population-Adjusted Treatment Comparisons

    Phillippo DM, Dias S, Ades AE, et al. Multilevel Network Meta-Regression for Population-Adjusted Treatment Comparisons. Journal of the Royal Statistical Society Series A: Statistics in Society 2020;183:1189–210

    work page 2020

  12. [12]

    Statistics in Medicine 2012;31:3840–57

    DoneganS,WilliamsonP,D’AlessandroU,andSmithCT.AssessingtheConsistencyAssumptionbyExploringTreatment by Covariate Interactions in Mixed Treatment Comparison Meta-Analysis: Individual Patient-Level Covariates versus Aggregate Trial-Level Covariates. Statistics in Medicine 2012;31:3840–57

    work page 2012

  13. [13]

    Meta-Analysis of Continuous Outcomes Combining Individual Patient Data and Aggregate Data

    Riley RD, Lambert PC, Staessen JA, et al. Meta-Analysis of Continuous Outcomes Combining Individual Patient Data and Aggregate Data. Statistics in Medicine 2008;27:1870–93

    work page 2008

  14. [14]

    Individual Patient- versus Group-Level Data Meta- RegressionsfortheInvestigationofTreatmentEffectModifiers:EcologicalBiasRearsItsUglyHead.StatisticsinMedicine 2002;21:371–87

    Berlin JA, Santanna J, Schmid CH, Szczech LA, and Feldman HI. Individual Patient- versus Group-Level Data Meta- RegressionsfortheInvestigationofTreatmentEffectModifiers:EcologicalBiasRearsItsUglyHead.StatisticsinMedicine 2002;21:371–87

    work page 2002

  15. [15]

    Meta-Regression Detected Associations between Heterogeneous Treatment Effects and Study-Level, but Not Patient-Level, Factors

    Schmid CH, Stark PC, Berlin JA, Landais P, and Lau J. Meta-Regression Detected Associations between Heterogeneous Treatment Effects and Study-Level, but Not Patient-Level, Factors. Journal of Clinical Epidemiology 2004;57:683–97

    work page 2004

  16. [16]

    Hua H, Burke DL, Crowther MJ, Ensor J, Tudur Smith C, and Riley RD. One-Stage Individual Participant Data Meta- Analysis Models: Estimation of Treatment-Covariate Interactions Must Avoid Ecological Bias by Separating out within- Trial and across-Trial Information. Statistics in Medicine 2017;36:772–89

    work page 2017

  17. [17]

    Two-Stage Bayesian Network Meta-Analysis of Individualized Treatment Rules for Multiple Treatments with Siloed Data

    Shen J, Moodie EE, and Golchi S. Two-Stage Bayesian Network Meta-Analysis of Individualized Treatment Rules for Multiple Treatments with Siloed Data. Statistical Methods in Medical Research 2025:09622802251387430

    work page 2025

  18. [18]

    Internet-Based Cognitive Behavioral Therapy for Depression: A Systematic Review and Individual Patient Data Network Meta-analysis

    Karyotaki E, Efthimiou O, Miguel C, et al. Internet-Based Cognitive Behavioral Therapy for Depression: A Systematic Review and Individual Patient Data Network Meta-analysis. JAMA Psychiatry 2021;78:361–71

    work page 2021

  19. [19]

    The Lancet Psychiatry 2021;8:500–11

    FurukawaTA,SuganumaA,OstinelliEG,etal.Dismantling,Optimising,andPersonalisingInternetCognitiveBehavioural TherapyforDepression:ASystematicReviewandComponentNetworkMeta-AnalysisUsingIndividualParticipantData. The Lancet Psychiatry 2021;8:500–11

    work page 2021