Pith. sign in

REVIEW 3 major objections 6 minor 30 references

Assessing the Impact of Covariate Distribution and Positivity Violation on Weighting-Based Indirect Comparisons: a Simulation Study

T0 review · 3 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read Under moderate positivity violations and non-normal covariates, the simplest moments-only MAIC estimator stays unbiased while full-data propensity score weighting remains biased.

desk verdict Useful simulation evidence, but MAIC-1's robustness to positivity violation is partly an artifact of the linear outcome DGM; the abstract oversells it. read the letter →

arxiv 2507.12241 v2 pith:XLMF6XP2 submitted 2025-07-16 stat.ME stat.AP

Arnaud Serret-Larmande , Jérôme Lambert , Stéphane Gaudry , David Hajage This is my paper
classification stat.MEstat.AP MSC 62P10
keywords indirecttreatmentcomparisonsmatching-adjustedcomparisonpropensityscoreweightingpositivityviolationscovariateoverlapMonteCarlosimulationaverageeffectincontrolsanchoredandunanchored
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper asks whether the moments-only reweighting used in Matching-Adjusted Indirect Comparison (MAIC) can be trusted when the usual overlap/positivity assumption is violated and covariates are non-normal. Using Monte Carlo simulations comparing MAIC matching only the first moment (MAIC-1), MAIC matching first and second moments (MAIC-2), and a full-individual-patient-data propensity score weighting (PSW) benchmark, it claims that MAIC-1 remains unbiased under moderate positivity violations and non-normal (including bimodal) covariates, while PSW stays biased even with the correct adjustment model. It also claims that omitting a key confounder produces far more bias than the choice of weighting estimator, and that anchoring on a common comparator arm reduces sensitivity to poor overlap. If right, a simple moments-based aggregate-data method can outperform a method with full individual data for the average-treatment-effect-in-controls estimand under limited overlap, and correct confounder adjustment matters more than estimator sophistication.

What carries the argument

The central object is the MAIC weight vector obtained by the method of moments: weights $w_i = \exp\{\alpha^T (X_i - h(X_b))\}$ are chosen so that weighted covariate moments in the index trial match reported moments in the target trial—mean-only for MAIC-1, means and variances for MAIC-2—whereas PSW weights each subject by the fitted odds $e(X_i)/(1-e(X_i))$ from a logistic model of trial membership fit to full IPD. This machinery creates a controlled contrast: PSW tries to replicate the whole covariate distribution of the target population, while MAIC-1 only aligns first moments. Under the paper's linear outcome model, aligning the mean of the effect modifier is sufficient to transport the treatment arm mean, so MAIC-1 keeps stable weights exactly where PSW's full-distribution matching produces propensity scores near 1 and extreme weights.

What would settle it

Re-run the same DGM-2, DGM-4, DGM-6 and DGM-8 scenarios with a nonlinear outcome such as $Y = X_1^2 + 2X_2 + (1+2X_1)I(Z=A) + I(Z=B) + \epsilon$, or with a binary outcome and a logit link, under the same positivity violations; if MAIC-1's bias grows with the violation while PSW's does not, the robustness claim is confined to the linear setting. Additionally, increase the overlap violation severity (larger shifts in the trial assignment model) until MAIC-1's bias becomes visible, and record where that breakdown occurs.

Watch

Extended reading notes

Core claim

Under the paper's data-generating scenarios (a continuous linear outcome with one treatment-effect modifier and one prognostic factor), MAIC-1 and MAIC-2 produced unbiased estimates of the treatment effect under moderate positivity violations in all four violation scenarios (normal, lognormal, and two bimodal distributions), whereas PSW, which has access to individual patient data from both trials and uses the correctly specified propensity score model, remained biased. The paper attributes this to the method of moments: matching only the first moment yields fewer extreme weights than full-distribution balancing, so MAIC-1's post-weighting propensity scores stay away from 1. Across all estimators, the largest source of bias was model misspecification, not the choice among MAIC-1, MAIC-2, and PSW. In the empirical AKIKI/AKIKI-2 application, MAIC-1 produced narrower confidence intervals under a positivity violation, consistent with the simulations.

Load-bearing premise

The simulation's outcome model is linear in the covariates, with $X_1$ as the only treatment-effect modifier; MAIC-1's unbiasedness under positivity violation rests on first-moment matching being enough to transport a linear mean, so the central claim could fail for nonlinear, non-collapsible, or time-to-event outcomes.

Editorial extensions

If this is right

  • MAIC-1 can be used cautiously for unanchored indirect comparisons under moderate overlap violations and non-normal covariates, where full-IPD PSW targeting the same average-treatment-effect-in-controls estimand would be biased.
  • Correctly identifying and including all treatment-effect modifiers and prognostic factors matters more than whether the analyst uses a moments-based or a full-propensity weighting estimator.
  • Anchored comparisons, which require balancing only effect modifiers, are less sensitive to positivity violations than unanchored ones and should be preferred when a common comparator arm exists.
  • Full access to individual patient data does not by itself protect against positivity failure for an ATC estimand; its practical value lies in analytical flexibility, such as choosing other estimands or trimming.
  • MAIC estimators can fail to converge under extreme overlap problems (up to 294/2000 iterations for unanchored MAIC-2 in DGM-6), so convergence diagnostics are needed in applied work.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Because the outcome model is linear, MAIC-1's robustness here is plausibly a property of moment-matching sufficiency for the ATC mean rather than of aggregate-versus-IPD data access; a natural test is whether MAIC-1 matches PSW when outcomes are nonlinear.
  • The findings suggest a practical heuristic: when post-weighting propensity scores under PSW concentrate near 1, a moments-only estimator may give a more stable estimate, but at the cost of balancing only means rather than the full distribution.
  • One could extend the comparison to overlap weights or trimmed PSW, which target estimands that avoid extreme weights; those methods may erase MAIC-1's apparent advantage while changing the estimand.
  • The real-world AKIKI application hints that alignment-date decisions can create positivity violations, so analysts should examine overlap before choosing an estimator, since the choice of follow-up start date can dominate the choice of weighting method.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. This manuscript reports a Monte Carlo simulation study that compares three weighting estimators for population-adjusted indirect comparisons targeting the average treatment effect in the trial-b population: MAIC matching first moments (MAIC-1), MAIC matching first and second moments (MAIC-2), and full-IPD propensity score weighting (PSW). Eight data-generating mechanisms vary the covariate distribution (normal, lognormal, bimodal) and the presence of positivity violations, under a continuous linear outcome model with X1 as the only treatment-effect modifier. The study evaluates bias, RMSE, variability ratio, and 95% coverage for anchored and unanchored versions, and also compares adjustment sets that include or omit key confounders. The authors report that MAIC-1 remains unbiased under moderate positivity violations, that MAIC-2 and PSW are more sensitive, and that model misspecification produces larger bias than estimator choice. They illustrate the methods on the AKIKI and AKIKI-2 trials of renal replacement therapy strategies.

Significance. If the results hold, the paper provides a useful comparison of MAIC against a full-IPD PSW estimator targeting the same ATC estimand, covering overlap violations and non-normal covariate distributions that are common in health-technology-assessment submissions. The simulation designs are clearly specified, the Monte Carlo setup is transparent, and the authors are explicit about many limitations in the Discussion. The finding that confounder selection matters more than estimator choice is consistent with previous work and is a useful practical message. However, the headline claim about MAIC-1's robustness to positivity violations is currently supported only under a linear, first-moment-sufficient outcome model; the paper's own Discussion acknowledges that non-collapsible and time-to-event outcomes are not investigated. The empirical application uses time-to-event outcomes without describing the corresponding methodology, which further limits the strength of the conclusions.

major comments (3)
  1. [Section 3.1, DGM-6] The central claim that MAIC-1 is unbiased under positivity violations is not established beyond the linear outcome model used in the simulations. The DGM in Section 2.3 sets Y = X1 + 2X2 + (1 + 2X1)I(Z=A) + I(Z=B) + epsilon, so E[Y(A)|X] = 1 + 3X1 + 2X2, which depends only on the first moments of X1 and X2. Because MAIC-1 matches first moments by construction, it is exactly unbiased for the transported A-arm mean even when overlap is poor; this is a property of the outcome model, not of the MAIC method itself. The abstract's statement that MAIC-1 shows 'resilience across moderate violations of assumptions' therefore overgeneralizes. The Discussion already acknowledges the limitation, but the abstract and Section 3.1 do not carry the qualifier. Please either temper the headline claim to 'under the linear outcome models considered here' or add simulations with nonlinear conditional means (e.g., quadratic terms, binary outcomes, or time-to-event outcomes) to demonstrate robustness beyond first-moment sufficiency.
  2. [Section 3.4] The handling of non-converged MAIC-2 iterations is not specified, which is load-bearing for the comparison between estimators. The paper reports that MAIC-2 failed to converge or produced absurd estimates in DGM-6 on 73/2000 anchored and 294/2000 unanchored iterations (Figures 5 and 6). If these iterations are excluded from the performance metrics, the reported bias, RMSE, and coverage for MAIC-2 are conditional on convergence, and this can itself induce selection bias. The authors should state exactly how non-converged iterations were treated, and should provide a sensitivity analysis (e.g., including all iterations, or using capped weights) to show that the qualitative conclusion that MAIC-2 is more sensitive to positivity violations is robust.
  3. [Section 3.4] The empirical application uses time-to-event outcomes (log(HR) in Figure 10) but the simulation study is restricted to continuous linear outcomes, and the manuscript does not describe how MAIC and PSW weights were applied to survival data (e.g., weighted Cox regression, bootstrap variance estimation that re-estimates weights, or handling of censoring). The Discussion explicitly states that time-to-event settings 'remain to be investigated,' which creates an inconsistency with the empirical analysis that reports hazard ratios. Please specify the empirical methods in detail, or reframe the AKIKI analysis as an illustrative extension with clearly stated untested assumptions.
minor comments (6)
  1. [Equation (3)] Equation (3) contains a typesetting error: the expression for the odds should read something like Pr(Ti = b | h(Xi)) / Pr(Ti = a | h(Xi)) = e(Xi)/(1 - e(Xi)); the current text has a misplaced equality and an opening parenthesis that is not closed.
  2. [Section 2.4] The sample-size description is confusing: 'two trials of 250 patients per arm' is later paraphrased as '500 patients per trial in the unanchored setting and 1,000 per trial in the anchored one.' Please clarify the effective sample sizes used by each estimator in each anchor setting.
  3. [Section 2.2.4] The variance computation is described only as a sum of arm-specific variances estimated by non-parametric bootstrap. It is unclear whether the bootstrap procedure re-estimates the MAIC or PSW weights in each resample, which is important for coverage properties. Please state the bootstrap procedure explicitly.
  4. [Figure 5 and Figure 6 captions] The captions report the number of non-converged iterations but do not define what 'did not converge' means (e.g., failure of the optimization algorithm, extreme weight values, or effective sample size below a threshold). Please provide a precise definition in the methods or figure captions.
  5. [Section 1] The introduction states that no prior study has compared MAIC with full-IPD PSW for the ATC estimand. In light of existing simulation-based work on population-adjusted indirect comparisons (e.g., Remiro-Azócar et al. 2021), please situate the novelty claim more carefully and clarify what specifically has not been done before.
  6. [Throughout] There are minor typographical issues, including 'Additionaly' in Section 1, inconsistent capitalization in 'Methods' versus 'methods', and some sentences in Section 3.1 where 'that is' is used loosely. A careful proofreading pass is recommended.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the simulation targets are externally generated and the linear DGM is an explicitly disclosed modeling scope, not an input relabeled as a prediction.

full rationale

This is a Monte Carlo simulation study in which the three estimators are evaluated against bias, RMSE, variability ratio, and coverage computed from independent simulation draws under explicit DGMs. No fitted parameter is renamed as a prediction, and no estimator equation is used as its own benchmark. MAIC-1's robustness under positivity violations does follow from the linear outcome model in Section 2.3, where first-moment matching is sufficient to transport the A-arm mean, but this is a transparently stated modeling assumption rather than a circular derivation: the paper does not claim to have derived the robustness result from the estimator's definition, and its Discussion explicitly limits the finding to 'a continuous outcome and a linear model' and notes that non-collapsible or time-to-event settings 'remain to be investigated.' The only self-citation (reference [22], the authors' own methodological review) supports a descriptive popularity claim ('MAIC remains the most widely used PAIC estimator') and is not load-bearing for the simulation results. No self-definitional step, fitted-input-as-prediction, or imported-uniqueness argument appears. The skeptical concern about overgeneralization beyond the linear DGM is a correctness/external-validity issue, not a circularity issue.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

The simulation is self-contained: all parameters are design choices in the DGMs, and the only external inputs are statistical assumptions and the AKIKI trial data. The free parameters are the hand-chosen coefficients and distribution shapes that define the eight scenarios. No new entities are postulated.

free parameters (6)
  • Outcome model coefficients = 1 for X1, 2 for X2, 1+2X1 for A arm, 1 for B arm, residual variance 1
    Hand-set in Section 2.3; define the linear outcome model under which first-moment matching is sufficient for unbiased transport.
  • Trial assignment logit coefficients = Vary by DGM, e.g. beta0 + X1 - 0.5X1^2 + X2 - 0.5X2^2 (DGM-1), 2beta0 + 2X1 + 2X2 (DGM-3/4), 0.5beta0 + 0.5X1…
    Hand-set in Section 2.3 to generate the intended overlap or positivity-violation regimes.
  • Beta0 (trial assignment intercept) = Calibrated by root-finding so marginal P(T=b)=0.5
    A free calibration constant; its value depends on the chosen coefficients and covariate distributions.
  • Delta sign for trial assignment = 1 or -1 depending on DGM (DGM-2, 4, 6, 8 use -1)
    Hand-set to toggle positivity violations; this is the key scenario-defining parameter.
  • Covariate distribution parameters = N(0,1); LogNormal(0,0.5) truncated at 5; mixture N(0,0.5)/N(3,0.5) with p=0.5; mixture N(0,1)/N(3,1) with p=0.5
    Hand-set to create normal, lognormal, and bimodal covariate scenarios.
  • Monte Carlo size = 250 patients per arm, 2000 iterations, 2000 bootstrap resamples
    Chosen in Sections 2.4 and 2.2.4; affects the precision and runtime of the performance metrics.
assumptions (5)
  • domain assumption Consistency, exchangeability, and no unmeasured confounding hold in the simulation DGM
    The DAG and outcome model in Section 2.3 make all confounders (X1, X2) measured and treatment assignment within trials randomized, so the weighting estimators target the defined causal estimand.
  • ad hoc to paper The true outcome model is linear in X1 and X2 with X1 as the only treatment-effect modifier
    The outcome equation Y = X1 + 2X2 + (1+2X1)I(Z=A) + I(Z=B) + epsilon in Section 2.3 is a hand-chosen simulation assumption; it is the main reason MAIC-1's mean matching succeeds.
  • domain assumption The ATC estimand requires only one-sided overlap (b support within a support)
    Section 2.1 defines positivity for the target population as the b covariate distribution being represented in a; this shapes how positivity violations are generated and interpreted.
  • domain assumption The trial-assignment model used by PSW is correctly specified in the main analysis
    Section 3.1 reports results under the correct trial assignment model; misspecification is examined separately, so the headline comparison assumes correct PSW specification.
  • standard math Bootstrap variance and normal-based intervals are adequate for inference
    Section 2.2.4 uses 2000 bootstrap iterations and 1.96-based coverage intervals; this assumes bootstrap consistency for the weighted estimators, which is standard but not proved here.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Assessing the Impact of Covariate Distribution and Positivity Violation on Weighting-Based Indirect Comparisons: a Simulation Study." pith.science (2026). https://pith.science/paper/XLMF6XP2

@misc{pith2026250712241,
  author       = {Pith},
  title        = {Pith review of: Assessing the Impact of Covariate Distribution and Positivity Violation on Weighting-Based Indirect Comparisons: a Simulation Study},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XLMF6XP2}},
  note         = {Machine review of arXiv:2507.12241}
}
read the original abstract

Population-Adjusted Indirect Comparisons (PAICs) are used to estimate treatment effects when direct comparisons are infeasible and individual patient data (IPD) are only available for one trial. Among PAIC methods, Matching-Adjusted Indirect Comparison (MAIC) is the most widely used. However, little is known about how MAIC performs under challenging conditions such as limited covariate overlap or markedly non-normal covariate distributions. We conducted a Monte Carlo simulation study comparing three estimators: (i) MAIC matching first moment (MAIC-1), (ii) MAIC matching first and second moments (MAIC-2), and (iii) a benchmark method leveraging full IPD -- Propensity Score Weighting (PSW). We examined eight scenarios ranging from ideal conditions to situations with positivity violations and non-normal (including bimodal) covariate distributions. We assessed both anchored and unanchored estimators and examined the impact of adjustment model misspecification. We also applied these estimators to real-world data from the AKIKI and AKIKI-2 trials, comparing renal replacement therapy strategies in critically ill patients. MAIC-1 demonstrated robust performance, remaining unbiased in the presence of moderate positivity violations and non-normal covariates, while MAIC-2 and PSW appeared more sensitive to positivity violations. All methods showed substantial bias when key confounders were omitted, emphasizing the importance of correct model specification. In real-world data, a consistent trend was found with MAIC-1 showing narrower confidence intervals with positivity violation. Our findings support the cautious use of unanchored MAICs and highlight MAIC-1's resilience across moderate violations of assumptions. However, the method's limited flexibility underscores the need for careful use in real-world settings.

Figures

Figures reproduced from arXiv: 2507.12241 by the authors.

Figure 1
Figure 1. Relationships among X1, X2, T, Z, and Y . 13 [PITH_FULL_IMAGE:figures/full_fig_p013_1.png] view at source ↗
Figure 2
Figure 2. Covariate distributions depending on DGM [PITH_FULL_IMAGE:figures/full_fig_p014_2.png] view at source ↗
Figure 3
Figure 3. Treatment effect estimation under DGM-1 and DGM-2 [PITH_FULL_IMAGE:figures/full_fig_p015_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Treatment effect estimation under DGM-3 and DGM-4 [PITH_FULL_IMAGE:figures/full_fig_p016_4.png]
Figure 5
Figure 5. Figure 5: Treatment effect estimation under DGM-5 and DGM-6 (asterisks indicate that estimators did not converge for [PITH_FULL_IMAGE:figures/full_fig_p017_5.png]
Figure 6
Figure 6. Figure 6: Treatment effect estimation under DGM-7 and DGM-8 (asterisks indicate that estimators did not converge for [PITH_FULL_IMAGE:figures/full_fig_p018_6.png]
Figure 7
Figure 7. Figure 7: Anchored estimators: X1 distribution before and after weighting + individual PS distribution depending on X1 value 19 [PITH_FULL_IMAGE:figures/full_fig_p019_7.png]
Figure 8
Figure 8. Figure 8: Bias of estimators depending on adequate model specification (the asterisk indicates that 3 iterations did not [PITH_FULL_IMAGE:figures/full_fig_p020_8.png]
Figure 9
Figure 9. Figure 9: AKIKI trials: treatment arms according to the RRT initiation time [PITH_FULL_IMAGE:figures/full_fig_p021_9.png]
Figure 10
Figure 10. Figure 10: Empirical data application on AKIKI RCTs: log(HR) estimate (95% CI) for the "more-delayed" strategy vs [PITH_FULL_IMAGE:figures/full_fig_p022_10.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

30 extracted references · 17 canonical work pages

  1. [1]

    Angelo de Claro, Lola Fashoyin-Aje, Nicole Gormley, Tamy Kim, Steven Lemery, Gautam U

    Sundeep Agrawal, Shaily Arora, Laleh Amiri-Kordestani, R. Angelo de Claro, Lola Fashoyin-Aje, Nicole Gormley, Tamy Kim, Steven Lemery, Gautam U. Mehta, Emma C. Scott, Harpreet Singh, Shenghui Tang, Marc R. Theoret, Richard Pazdur, Paul G. Kluetz, and Julia A. Beaver. Use of Single-Arm Trials for US Food and Drug Administration Drug Approval in Oncology, 2...

    work page arXiv 2002

  2. [2]

    William H. Asquith. L-moments and TL-moments of the generalized lambda distribution.Computational Statistics & Data Analysis, 51(9):4484–4496, May 2007. ISSN 0167-9473. doi: 10.1016/j.csda.2006.07.016

    work page doi:10.1016/j.csda.2006.07.016 2007

  3. [3]

    Peter C. Austin. An Introduction to Propensity Score Methods for Reducing the Effects of Confounding in Observational Studies. Multivariate Behavioral Research, 46(3):399–424, May 2011. ISSN 1532-7906. doi: 10.1080/00273171.2011.568786

    arXiv 2011

  4. [4]

    Beaver, Lynn J

    Julia A. Beaver, Lynn J. Howie, Lorraine Pelosof, Tamy Kim, Jinzhong Liu, Kirsten B. Goldberg, Rajeshwari Sridhara, Gideon M. Blumenthal, Ann T. Farrell, Patricia Keegan, Richard Pazdur, and Paul G. Kluetz. A 25-Year Experience of US Food and Drug Administration Accelerated Approval of Malignant Hematology and Oncology Drugs and Biologics: A Review. JAMA ...

    work page arXiv 2018

  5. [5]

    Bucher, Gordon H

    Heiner C. Bucher, Gordon H. Guyatt, Lauren E. Griffith, and Stephen D. Walter. The results of direct and indirect treatment comparisons in meta-analysis of randomized controlled trials. Journal of Clinical Epidemiology, 50(6): 683–691, June 1997. ISSN 0895-4356. doi: 10.1016/S0895-4356(97)00049-8

    doi:10.1016/s0895-4356(97)00049-8 1997

  6. [6]

    Park, Jeroen P

    Harlan Campbell, Julie E. Park, Jeroen P. Jansen, and Shannon Cope. Estimating marginal treatment effects from observational studies and indirect treatment comparisons: When are standardization-based methods preferable to those based on propensity score weighting?, October 2023

    work page 2023

  7. [7]

    Jaime Caro and K

    J. Jaime Caro and K. Jack Ishak. No Head-to-Head Trial? Simulate the Missing Arms:. PharmacoEconomics, 28 (10):957–967, October 2010. ISSN 1170-7690. doi: 10.2165/11537420-000000000-00000

    doi:10.2165/11537420-000000000-00000 2010

  8. [8]

    Matching-Adjusted Indirect Comparison Studies in Oncology: A Scoping Review Focused on Reporting Quality

    Cecília Menezes Farinasso, Vinícius Lins Ferreira, Flávia Cordeiro Medeiros, Aline Pereira da Rocha, Patrícia do Carmo Silva Parreira, Layssa Andrade Oliveira, Lays Pires Marra, Rosa Camila Lucchetta, and Haliton Alves de Oliveira. Matching-Adjusted Indirect Comparison Studies in Oncology: A Scoping Review Focused on Reporting Quality. Value in Health Reg...

    work page doi:10.1016/j.vhri.2025 2025

Show all 30 references
  1. [9]

    The Effects of Model Misspecification in Unanchored Matching-Adjusted Indirect Comparison: Results of a Simulation Study

    Anthony James Hatswell, Nick Freemantle, and Gianluca Baio. The Effects of Model Misspecification in Unanchored Matching-Adjusted Indirect Comparison: Results of a Simulation Study. Value in Health, 23(6): 751–759, June 2020. ISSN 10983015. doi: 10.1016/j.jval.2020.02.008

    2020 doi

  2. [10]

    Causal Inference: What If

    Miguel A Hernán and James M Robins. Causal Inference: What If. CRC Press, 2020

    2020

  3. [11]

    Samjoo, and Chris Cameron

    Ataru Igarashi, Shiro Tanaka, Raf De Moor, Nan Li, Mariko Hirozane, David Bin-Chia Wu, Li Wen Hong, Dae Young Yu, Mahmoud Hashim, Brian Hutton, Krista Tantakoun, Christopher Olsen, Fatemeh Mirzayeh Fashami, Imtiaz A. Samjoo, and Chris Cameron. Indirect Treatment Comparisons in...

    2021

  4. [12]

    Thomas, and Fan Li

    Fan Li, Laine E. Thomas, and Fan Li. Addressing Extreme Propensity Scores via the Overlap Weights. American Journal of Epidemiology, 188(1):250–257, January 2019. ISSN 0002-9262. doi: 10.1093/aje/kwy201

    2019 doi

  5. [13]

    Considerations for Single-Arm Trials to Support Accelerated Approval of Oncology Drugs, November 2024

    Feinan Lu, Tao Wang, Ying Lu, and Jie Chen. Considerations for Single-Arm Trials to Support Accelerated Approval of Oncology Drugs, November 2024

    2024

  6. [14]

    The Acceptance of Indirect Treatment Comparison Methods in Oncology by Health Technology Assessment Agencies in England, France, Germany, Italy, and Spain

    Bérengère Macabeo, Théophile Rotrou, Aurélie Millier, Clément François, and Philippe Laramée. The Acceptance of Indirect Treatment Comparison Methods in Oncology by Health Technology Assessment Agencies in England, France, Germany, Italy, and Spain. PharmacoEconomics - Open, 8...

    2024 doi

  7. [15]

    Park, Harlan Campbell, Kevin Towle, Yong Yuan, Jeroen P

    Julie E. Park, Harlan Campbell, Kevin Towle, Yong Yuan, Jeroen P. Jansen, David Phillippo, and Shannon Cope. Unanchored Population-Adjusted Indirect Comparison Methods for Time-to-Event Outcomes Using Inverse Odds Weighting, Regression Adjustment, and Doubly Robust Methods Wit...

    2024 doi

  8. [16]

    Phillippo, Anthony E

    David M. Phillippo, Anthony E. Ades, Sofia Dias, Stephen Palmer, Keith R. Abrams, and Nicky J. Welton. Methods for Population-Adjusted Indirect Comparisons in Health Technology Appraisal. Medical Decision Making, 38(2):200–211, February 2018. ISSN 0272-989X, 1552-681X. doi: 10...

    2018 doi

  9. [17]

    Phillippo, Sofia Dias, A

    David M. Phillippo, Sofia Dias, A. E. Ades, and Nicky J. Welton. Multilevel network meta-regression for general likelihoods: Synthesis of individual and aggregate data with applications to survival analysis, January 2024

    2024

  10. [18]

    DM Phillippo, A. E. Ades, Sofia Dias, Stephen Palmer, Keith R. Abrams, and Nicky J. Welton. Methods for Population-Adjusted Indirect Comparisons in Submission to NICE. Technical Report 18, Decision Support Unit, National Institute for Health and Care Excellence, University of ...

    2016

  11. [19]

    Target estimands for population-adjusted indirect comparisons

    Antonio Remiro-Azócar. Target estimands for population-adjusted indirect comparisons. Statistics in Medicine, 41(28):5558–5569, December 2022. ISSN 0277-6715, 1097-0258. doi: 10.1002/sim.9413

    2022 doi

  12. [20]

    Methods for population adjustment with limited access to individual patient data: A review and simulation study

    Antonio Remiro-Azócar, Anna Heath, and Gianluca Baio. Methods for population adjustment with limited access to individual patient data: A review and simulation study. Research Synthesis Methods, 12(6):750–775, 2021. ISSN 1759-2887. doi: 10.1002/jrsm.1511

    2021 doi

  13. [21]

    target estimands for population-adjusted indirect comparisons

    Estelle Russek-Cohen. Discussion of “target estimands for population-adjusted indirect comparisons” by Antonio Remiro-Azocar. Statistics in Medicine, 41(28):5573–5576, 2022. ISSN 1097-0258. doi: 10.1002/sim.9533

    2022 doi

  14. [22]

    A method- ological review of population-adjusted indirect comparisons reveals inconsistent reporting and suggests publi- cation bias

    Arnaud Serret-Larmande, Belkacem Zenati, Agnès Dechartres, Jérôme Lambert, and David Hajage. A method- ological review of population-adjusted indirect comparisons reveals inconsistent reporting and suggests publi- cation bias. Journal of Clinical Epidemiology, 163:1–10, Novemb...

    2023 doi

  15. [23]

    Signorovitch, Eric Q

    James E. Signorovitch, Eric Q. Wu, Andrew P. Yu, Charles M. Gerrits, Evan Kantor, Yanjun Bao, Shiraz R. Gupta, and Parvez M. Mulani. Comparative effectiveness without head-to-head trials: A method for matching-adjusted indirect comparisons applied to psoriasis treatment with a...

    2010 doi

  16. [24]

    Signorovitch J, Erder MH, Xie J, Sikirica V, Lu M, Hodgkins PS, and Wu EQ. Comparative effectiveness research using matching-adjusted indirect comparison: An application to treatment with guanfacine extended release or atomoxetine in children with attention-deficit/hyperactivi...

    2012

  17. [25]

    Tran, Tuan Anh Le, Thi Thu Pham, and Tat-Thang V o

    Bang Truong, Lan-Anh T. Tran, Tuan Anh Le, Thi Thu Pham, and Tat-Thang V o. Population adjusted-indirect comparisons in health technology assessment: A methodological systematic review. Research Synthesis Methods, 14(5):660–670, September 2023. ISSN 1759-2879, 1759-2887. doi: ...

    2023 doi

  18. [26]

    Rapid access to innovative medicinal products while ensuring relevant health technology assessment

    Antoine Vanier, Judith Fernandez, Sophie Kelley, Lise Alter, Patrick Semenzato, Corinne Alberti, Sylvie Chevret, Dominique Costagliola, Michel Cucherat, Bruno Falissard, François Gueyffier, Jérôme Lambert, Etienne Lengliné, Clara Locher, Florian Naudet, Raphael Porcher, Rodolp...

    2022

  19. [27]

    Matching-adjusted Indirect Comparisons vs Propensity Score Matching with Individual Patient-level Data to Estimate Treatment Efficacy

    Emily C L Wong, Parambir S Dulai, John K Marshall, Vipul Jairath, Walter Reinisch, and Neeraj Narula. Matching-adjusted Indirect Comparisons vs Propensity Score Matching with Individual Patient-level Data to Estimate Treatment Efficacy. Inflammatory Bowel Diseases, page izad07...

    2023 doi

  20. [28]

    Harmonic transform-based density estimation method for uncertainty propagation and reliability analysis involving multi-modal distributions

    Quanfu Yu and Jun Xu. Harmonic transform-based density estimation method for uncertainty propagation and reliability analysis involving multi-modal distributions. Mechanical Systems and Signal Processing, 190:110113, May 2023. ISSN 0888-3270. doi: 10.1016/j.ymssp.2023.110113

    2023

  21. [29]

    Moving toward best practice when using propensity score weighting in survey observational studies, January 2025

    Yukang Zeng, Fan Li, and Guangyu Tong. Moving toward best practice when using propensity score weighting in survey observational studies, January 2025

    2025

  22. [30]

    Hubbard, Jessica Chubak, Jason Roy, and Nandita Mitra

    Yaqian Zhu, Rebecca A. Hubbard, Jessica Chubak, Jason Roy, and Nandita Mitra. Core Concepts in Pharmacoepi- demiology: Violations of the Positivity Assumption in the Causal Analysis of Observational Data: Consequences and Statistical Approaches. Pharmacoepidemiology and drug s...

    2021 doi

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.