Pith. sign in

REVIEW 2 major objections 4 minor 39 references

Transporting results from a trial to an external target population when trial participation impacts adherence

T0 review · 2 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read This paper shows that when trial participation affects adherence, the target-population mean outcome under each treatment is identifiable from trial data once the user specifies the covariate-conditional ratio of adherence in the target…

desk verdict Useful sensitivity analysis for transporting trials when adherence differs, but the printed one-step estimator has a sign error that contradicts its own EIC derivation. read the letter →

arxiv 2506.00157 v1 pith:GFJM67BZ submitted 2025-05-30 stat.ME

Rachael K. Ross , Ivan Diaz , Amy J. Pitts , Elizabeth A. Stuart , Kara E. Rudolph This is my paper
classification stat.ME MSC 62D20
keywords transportabilitygeneralizabilityadherencesensitivityanalysistargetpopulationefficientinfluencefunctiondoublerobustnessone-stepestimator
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

Randomized trials can fail to generalize because trial activities—compensation, monitoring, visit frequency—can raise adherence above real-world levels, and adherence is a mediator of the treatment-outcome relationship. The paper shows that, if the researcher can specify the covariate-conditional ratio of adherence in the target population to adherence in the trial, $\delta_a(W)$, the target mean outcome under each treatment is identifiable from trial data by reweighting the trial's outcome model with the rescaled adherence probabilities. It provides a plug-in estimator and a one-step, efficient-influence-curve estimator that is double robust and compatible with machine-learning nuisance models. In an application transporting relapse risk for two opioid-use-disorder medications from a trial to a national treatment-admission cohort, the calibrated estimates raise the risk difference, suggesting the real-world benefit of buprenorphine-naloxone over extended-release naltrexone is larger than naive transport suggests. A reader should care because the method turns an unmeasured mediator gap that usually invalidates generalization into a transparent, user-controlled sensitivity analysis.

What carries the argument

The load-bearing object is the adherence ratio $\delta_a(W)=\Pr(Z^a=1|W,S=0)/\Pr(Z^a=1|W,S=1)$, a single user-specified multiplier that converts trial adherence into target adherence within each covariate stratum. It enters equation (3) through the term $m_a(W)\delta_a(W)$, effectively reweighting the trial's observed outcome mixture; the same ratio makes the target mean identifiable without any adherence data from the target. The other mechanism is the efficient influence curve one-step estimator, which corrects the plug-in estimate with a term involving outcome, adherence, treatment, and selection nuisance models, and provides double robustness under the model subsets listed in the appendix.

What would settle it

Measure adherence and outcomes in a sample from the target population, transport the trial results with the proposed formula, and compare the calibrated prediction to the observed target mean for each treatment: if the discrepancy exceeds sampling error for every plausible $\delta_a(W)$, the no-direct-effect assumption is false. A sharper version would randomize the intensity of trial-like activities in a target sample and test whether outcomes shift with the activities while holding adherence fixed.

Watch

Extended reading notes

Core claim

The paper's central claim is equation (3): under assumptions A1*–A4* and A5, $E[Y^a|S=0]$ equals $E[\,E(Y|A=a,Z=1,W,S=1)\,m_a(W)\delta_a(W) + E(Y|A=a,Z=0,W,S=1)(1-m_a(W)\delta_a(W))\,|\,S=0\,]$, where $m_a(W)=\Pr(Z=1|A=a,W,S=1)$ is trial adherence and $\delta_a(W)$ is the user-specified target-to-trial adherence ratio. When $\delta_a(W)=1$, the formula collapses to standard covariate-standardized transportability; values below 1 encode the belief that trial activities inflate adherence. Because $\delta_a(W)$ is unidentifiable from data, the paper frames the estimand as a sensitivity analysis and gives guidance for setting it by bounds or Monte Carlo draws. The accompanying one-step estimator is derived from the efficient influence curve and is shown to be consistent when any of three subsets of nuisance models is correctly specified, which supports valid inference with machine learning.

Load-bearing premise

The load-bearing premise is that participation in the trial changes outcomes only through adherence: if trial monitoring, provider contact, or compensation has any direct effect on the outcome, the identification formula fails and no user-specified adherence ratio can correct it.

Editorial extensions

If this is right

  • When $\delta_a(W)=1$, the proposed estimand reduces to the standard transported mean, so the method nests ordinary covariate-standardized transportability as a limiting case.
  • Researchers who specify a range for $\delta_a(W)$ obtain bounds on the target estimand, and those who specify a probability distribution can summarize the target estimand across Monte Carlo draws.
  • The one-step estimator remains consistent when only certain subsets of nuisance models are correct—outcome and adherence, or treatment and selection plus adherence, or outcome plus treatment and selection—so it supports machine-learning nuisance estimation with cross-fitting.
  • In the opioid-use-disorder application, lowering the adherence ratio from 1 to 0.5 raises the transported risk difference from 31.0 to 34.1 percentage points, and the Monte Carlo analysis with $\delta_1 \le \delta_0$ gives a median risk difference of 35.4 points.

Reading between the lines

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

  • The same ratio correction could be applied to any mediator of the treatment–outcome relationship that trial participation changes, such as visit attendance or care quality, provided the no-direct-effect assumption holds for the outcome.
  • Because the identification depends entirely on external specification of $\delta_a(W)$, the method would benefit from validation substudies where target adherence is measured; such data would let researchers estimate rather than assume the ratio.
  • The paper notes that multi-valued adherence would require one ratio per adherence category; an immediate extension would be to let those ratios be drawn from a correlated distribution to respect beliefs like 'initiation is harder for XR-NTX than BUP-NX.'
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

2 major / 4 minor

Summary. The paper proposes a sensitivity analysis for transporting treatment effects from a randomized trial to an external target population when trial participation affects adherence and adherence is unmeasured in the target. The central estimand is E[Y^a|S=0], identified under assumptions A1*–A4* and A5 as the covariate-standardized trial outcome in which the trial adherence probability m_a(W) is replaced by the user-specified scaled adherence δ_a(W)m_a(W). The paper provides a plug-in estimator and a one-step estimator claimed to be double robust and to support machine-learning nuisance estimation, discusses specification of δ_a(W), and applies the method to transport X:BOT relapse risks to a TEDS-A target. The identification argument is internally coherent under the stated no-direct-effect assumption, and the appendix contains a detailed EIC derivation. However, the printed one-step estimator in the main text and in the appendix estimating equations has a minus sign in the offset term where the EIC derivation requires a plus sign, so the estimator as written is not the estimator whose double-robustness properties are proved.

Significance. If the sign issue is corrected, the paper would make a useful contribution: it addresses a real gap in transportability methodology by allowing a transparent, user-specified sensitivity parameter for adherence differences, rather than assuming adherence is exchangeable across trial and target. The paper is unusually complete in shipping identification proofs, an EIC derivation, estimating equations, and code/data links, and the application to opioid use disorder treatment is relevant and well-motivated. The explicit scope condition that trial participation has no direct effect on outcome is stated clearly, and the reliance on external δ_a(W) is framed as sensitivity analysis, not as data-driven identification. The main weakness is the internal inconsistency in the one-step estimator, which jeopardizes the double-robustness and machine-learning inference claims as printed.

major comments (2)
  1. [One-step estimator (main text) and Appendix estimating equations] The printed one-step estimator contains the term −δ_a(W)(Q̂_{a,1}(W)−Q̂_{a,0}(W))(Z−m̂_a(W)) in the trial contribution, both in the main text and in the Appendix estimating equations. The EIC derived in Appendix 'Derivation of conjectured EIC' and the 'Conjectured EIC' used in the double-robustness proof instead have +δ_a(W)(Q̂_{a,1}(W)−Q̂_{a,0}(W))(Z−m̂_a(W)). The plus sign is the correct one: differentiating Q_1 m δ + Q_0(1−mδ) with respect to m yields +δ(Q_1−Q_0)(Z−m). With the printed minus sign, the estimator is not the plug-in estimator plus the mean EIC, and the extra discrepancy has conditional mean proportional to (1−h(W)) δ_a(W)(Q̂_{a,1}(W)−Q̂_{a,0}(W))(m_a(W)−m̂_a(W)), which vanishes only when the adherence model is correctly specified. Consequently, the claimed double-robustness subset {Q, g, h} with m misspecified does not hold for the estimator as written.
  2. [Application and Table 3] Because the application uses the one-step estimator and the printed formula contains the sign error, the point estimates, confidence intervals, and simulation intervals in Table 3 and Figure 4 are not reliable unless the accompanying code implements the correct plus-sign version. After correcting the sign in the main text and Appendix, please verify the code and regenerate any affected numerical results, or state explicitly that the code uses the correct EIC if that is the case.
minor comments (4)
  1. [Assumptions (main text)] Assumption A1 is written with a minus sign, 'E(Y^a|W,S=1)−E(Y^a|W,S=0)', where an equality is intended; please replace the minus with '='.
  2. [Appendix, 'Rate double-robustness'] The definition of µ near the start of the proof reads 'µ = Q1mδ − Q0(1−mδ)', but the estimand and all subsequent algebra use a plus sign; please correct this typo.
  3. [Table 3] Several confidence intervals in Table 3 have upper endpoints displayed as '1.0' (e.g., 84.9 (67.1, 1.0) and 90.1 (72.1, 1.0)); these appear to be placeholders and should be replaced with the actual upper bounds.
  4. [One-step estimator (main text)] The sentence 'This estimator includes four nuisance models, bQ_a,z(W), bm_a(W), bga(W) or bh(W)' should use 'and' rather than 'or', because the estimator as printed depends on both g and h.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the sensitivity parameter δ is an explicitly unidentifiable external input, and the identification and efficiency derivations are self-contained.

full rationale

The central identification result (equation 3) is derived in the Appendix under assumptions A1*--A4* and A5 by substituting the definition δ_a(W) = Pr(Z^a=1|W,S=0) / Pr(Z^a=1|W,S=1) (equation 1) into the law of total probability expansion. δ is not fitted to the target outcome and is explicitly acknowledged as unidentifiable: 'δa is unknown and cannot be identified with the data, so this parameter must be specified for estimation.' Thus the dependence of the estimand on δ is a designed sensitivity analysis rather than a prediction that reduces to its inputs. The one-step estimator is presented as a plug-in plus estimated efficient influence curve, with a full EIC derivation and remainder-term double-robustness decomposition in the Appendix; these proofs do not rely on self-citation. The only self-citations (references 6 and 12) are contextual or supporting and are not load-bearing: the asymptotic and inference claims are also anchored to external references [14] and to the paper's own Appendix derivations. A separate, non-circularity concern is that the printed one-step estimator and Appendix estimating equations display a minus sign before the δ(Q̂1−Q̂0)(Z−m̂) term, whereas the derived EIC has a plus sign; this is an internal consistency/correctness issue rather than an equivalence of estimand and input, so it does not raise the circularity score.

Assumptions & free parameters 2 free parameters · 7 assumptions · 0 invented entities

The method's central estimand depends on a user-specified adherence ratio delta, which is the key free input. The causal assumptions A1*-A5 are the identifying backbone. No new physical or conceptual entities are introduced.

free parameters (2)
  • delta_1 (XR-NTX adherence ratio) = 0.5 and 0.75 in static analyses; trapezoidal distribution with min 0.5, max 1, modes 0.6 and 0.75 in MC analysis
    User-specified ratio of target adherence to trial adherence for XR-NTX. Cannot be identified from the data; chosen from external belief that trial activities improve initiation more for XR-NTX.
  • delta_0 (BUP-NX adherence ratio) = 0.5 and 0.75 in static analyses; trapezoidal distribution with min 0.5, max 1, modes 0.75 and 0.9 in MC analysis
    User-specified ratio for BUP-NX. External belief that trial activities have less impact on BUP-NX adherence than on XR-NTX adherence.
assumptions (7)
  • domain assumption A1*: E(Y^a|W,Z^a,S=1)=E(Y^a|W,Z^a,S=0)
    Mean conditional exchangeability over samples given baseline covariates and potential adherence. Needed for identification in equation (3); unverifiable because Z^a is unobserved in the target.
  • domain assumption No direct effect of S on Y (no S->Y arrow)
    Stated after equation (1): trial activities affect the outcome only through adherence. If false, equation (3) is invalid.
  • domain assumption A3*: (Y^a,Z^a) independent of A given W,S=1
    Conditional exchangeability over treatment assignment in the trial; holds by randomization but requires that randomization is the only source of treatment assignment.
  • standard math A5: causal consistency in the trial
    Observed Z and Y equal potential values under assigned A. Standard consistency assumption.
  • domain assumption A2*/A4*: positivity of trial participation and treatment assignment conditional on W and Z^a
    Required for the outer expectations and the inverse-probability weights in the one-step estimator.
  • domain assumption In the application, TEDS-A admissions represent the target population and are treated as independent
    TEDS-A is admission-based, not patient-based; independence and representativeness are assumed for the transport analysis.
  • domain assumption Complete-case analysis assumes missingness does not induce selection bias
    8.3% of TEDS-A admissions with any missing covariate were excluded, implicitly assuming missing at random.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Transporting results from a trial to an external target population when trial participation impacts adherence." pith.science (2026). https://pith.science/paper/GFJM67BZ

@misc{pith2026250600157,
  author       = {Pith},
  title        = {Pith review of: Transporting results from a trial to an external target population when trial participation impacts adherence},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GFJM67BZ}},
  note         = {Machine review of arXiv:2506.00157}
}
read the original abstract

Randomized clinical trials are considered the gold standard for informing treatment guidelines, but results may not generalize to real-world populations. Generalizability is hindered by distributional differences in baseline covariates and treatment-outcome mediators. Approaches to address differences in covariates are well established, but approaches to address differences in mediators are more limited. Here we consider the setting where trial activities that differ from usual care settings (e.g., monetary compensation, follow-up visits frequency) affect treatment adherence. When treatment and adherence data are unavailable for the real-world target population, we cannot identify the mean outcome under a specific treatment assignment (i.e., mean potential outcome) in the target. Therefore, we propose a sensitivity analysis in which a parameter for the relative difference in adherence to a specific treatment between the trial and the target, possibly conditional on covariates, must be specified. We discuss options for specification of the sensitivity analysis parameter based on external knowledge including setting a range to estimate bounds or specifying a probability distribution from which to repeatedly draw parameter values (i.e., use Monte Carlo sampling). We introduce two estimators for the mean counterfactual outcome in the target that incorporates this sensitivity parameter, a plug-in estimator and a one-step estimator that is double robust and supports the use of machine learning for estimating nuisance models. Finally, we apply the proposed approach to the motivating application where we transport the risk of relapse under two different medications for the treatment of opioid use disorder from a trial to a real-world population.

Figures

Figures reproduced from arXiv: 2506.00157 by the authors.

Figure 1
Figure 1. Directed acyclic graphs (DAG) and Single-World Intervention Graphs (SWIG) i) DAG W S A Z Y SWIG W S A a Z a Y a ii) DAG W S A Z Y SWIG W S A a Z a Y a 21 [PITH_FULL_IMAGE:figures/full_fig_p021_1.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

39 extracted references · 30 canonical work pages

  1. [1]

    E., Russell, M., Luo, S

    Rudolph, K. E., Russell, M., Luo, S. X., Rotrosen, J. & Nunes, E. V. Under-representation of key demographic groups in opioid use disorder trials. en.Drug and Alcohol Dependence Reports4, 100084. issn: 27727246. https : / / linkinghub . elsevier . com / retrieve / pii / S2772724622000592(2022) (Sept. 2022)

    work page 2022

  2. [2]

    & Bareinboim, E.Transportability Across Studies: A Formal Approachen

    Pearl, J. & Bareinboim, E.Transportability Across Studies: A Formal Approachen. Tech. rep. (Defense Technical Information Center, Fort Belvoir, VA, Mar. 2011).http://www.dtic.mil/docs/citations/ ADA557437 (2023)

    work page 2023

  3. [3]

    Cole, S. R. & Stuart, E. A. Generalizing Evidence From Randomized Clinical Trials to Target Popula- tions, The ACTG 320 Trial.American Journal of Epidemiology172, 107–115 (2010)

    work page 2010

  4. [4]

    Lesko, C. R. et al. Generalizing Study Results: A Potential Outcomes Perspective.Epidemiology 28. ISBN: 0000000000000, 553–561.issn: 1044-3983 (2017)

    work page 2017

  5. [5]

    J., Robertson, S

    Dahabreh, I. J., Robertson, S. E., Steingrimsson, J. A., Stuart, E. A. & Hernán, M. A. Extending inferences from a randomized trial to a new target population. en.Statistics in Medicine 39, 1999–

    work page 1999

  6. [6]

    Rudolph, K. E. & van der Laan, M. J. Robust estimation of encouragement design intervention effects transported across sites. en.Journal of the Royal Statistical Society: Series B (Statistical Methodology) 79, 1509–1525. issn: 13697412. https : / / onlinelibrary . wiley . com / doi / 10 . 1111 / rssb . 12213 (2022) (Nov. 2017)

    work page 2022

  7. [7]

    J., Robertson, S

    Dahabreh, I. J., Robertson, S. E. & Hernán, M. A.Generalizing and transporting inferences about the effects of treatment assignment subject to non-adherenceVersion Number: 1. 2022.https://arxiv. org/abs/2211.04876 (2024)

    arXiv 2024

  8. [8]

    The calibration of treatment effects from clinical trials to target populations

    Frangakis, C. The calibration of treatment effects from clinical trials to target populations. en.Clinical Trials 6, 136–140. issn: 1740-7745, 1740-7753. https : / / journals . sagepub . com / doi / 10 . 1177 / 1740774509103868 (2025) (Apr. 2009)

    work page 2025

Show all 39 references
  1. [9]

    Wang, S. V. et al.Emulation of Randomized Clinical Trials With Nonrandomized Database Analyses: Results of 32 Clinical Trials. en.JAMA 329, 1376. issn: 0098-7484. https : / / jamanetwork . com / journals/jama/fullarticle/2804067 (2025) (Apr. 2023)

    2025

  2. [10]

    M.et al.Emulating Randomized Clinical Trials With Nonrandomized Real-World Evidence Studies: First Results From the RCT DUPLICATE Initiative

    Franklin, J. M.et al.Emulating Randomized Clinical Trials With Nonrandomized Real-World Evidence Studies: First Results From the RCT DUPLICATE Initiative. en.Circulation 143, 1002–1013. issn: 0009-7322, 1524-4539. https://www.ahajournals.org/doi/10.1161/CIRCULATIONAHA.120.0517...

    2025 doi

  3. [11]

    Dahabreh, I. J. & Bibbins-Domingo, K. Causal Inference About the Effects of Interventions From Ob- servational Studies in Medical Journals. en.JAMA331, 1845. issn: 0098-7484.https://jamanetwork. com/journals/jama/fullarticle/2818746 (2025) (June 2024)

    2025

  4. [12]

    E., Díaz, I

    Renson, A., Montoya, L., Goin, D. E., Díaz, I. & Ross, R. K.Pulling back the curtain: the road from statistical estimand to machine-learning based estimator for epidemiologists (no wizard required)en. arXiv:2502.05363 [stat]. Feb. 2025.http://arxiv.org/abs/2502.05363 (2025)

    2025 arXiv

  5. [13]

    Semiparametric doubly robust targeted double machine learning: a reviewen.arXiv:2203.06469 [stat]

    Kennedy,E.H. Semiparametric doubly robust targeted double machine learning: a reviewen.arXiv:2203.06469 [stat]. Jan. 2023.http://arxiv.org/abs/2203.06469 (2023)

    2023 arXiv

  6. [14]

    & Vansteelandt, S

    Hines, O., Dukes, O., Diaz-Ordaz, K. & Vansteelandt, S. Demystifying statistical learning based on efficient influence functions. en.The American Statistician76. arXiv:2107.00681 [math, stat], 292–304. issn: 0003-1305, 1537-2731.http://arxiv.org/abs/2107.00681 (2023) (July 2022)

    2023 arXiv

  7. [15]

    et al.)doi.wiley.com/10.1002/9781118445112.stat08068, 1–14 (John Wiley & Sons, Ltd, Chichester, UK, May 2018).isbn: 978-1-118-44511-2

    Daniel,R.M.en.in Wiley StatsRef: Statistics Reference Online(edsBalakrishnan,N. et al.)doi.wiley.com/10.1002/9781118445112.stat08068, 1–14 (John Wiley & Sons, Ltd, Chichester, UK, May 2018).isbn: 978-1-118-44511-2

    2018 doi

  8. [16]

    & van der Laan, M

    Tran, L., Petersen, M., Schwab, J. & van der Laan, M. J.Robust variance estimation and inference for causal effect estimationen. arXiv:1810.03030 [math, stat]. Oct. 2018.http://arxiv.org/abs/1810. 03030 (2024). 15

    2024 arXiv

  9. [17]

    Manski, C. F. Nonparametric Bounds on Treatment Effects. en.The American Economic Review80, 6 (1990)

    1990

  10. [18]

    L., Fox, M

    Lash, T. L., Fox, M. P. & Fink, A. K. Applying Quantitative Bias Analysis to Epidemiologic Data (Springer, New York, NY, 2009)

    2009

  11. [19]

    et al.Comparativeeffectivenessofextended-releasenaltrexoneversusbuprenorphine-naloxone for opioid relapse prevention (X:BOT): a multicentre, open-label, randomised controlled trial

    Lee,J.D. et al.Comparativeeffectivenessofextended-releasenaltrexoneversusbuprenorphine-naloxone for opioid relapse prevention (X:BOT): a multicentre, open-label, randomised controlled trial. en. The Lancet 391, 309–318. issn: 01406736. https : / / linkinghub . elsevier . com /...

    2023

  12. [20]

    Treatment Episode Data Set (TEDS)https : / / www . samhsa . gov / data / data - we - collect / teds - treatment-episode-data-set (2025)

    2025

  13. [21]

    TIP 63: Medications for Opioid Use Disordertech. rep. (Subtance Abuse and Mental Health Services Administration, July 2021). https : / / store . samhsa . gov / product / TIP - 63 - Medications - for - Opioid-Use-Disorder-Full-Document/PEP21-02-01-002 (2023)

    2023

  14. [22]

    et al.Rapid Initiation of Injection Naltrexone for Opioid Use Disorder: A Stepped-Wedge Cluster Randomized Clinical Trial

    Shulman, M. et al.Rapid Initiation of Injection Naltrexone for Opioid Use Disorder: A Stepped-Wedge Cluster Randomized Clinical Trial. en. JAMA Network Open 7, e249744. issn: 2574-3805. https : //jamanetwork.com/journals/jamanetworkopen/fullarticle/2818410 (2024) (May 2024)

    2024

  15. [23]

    T.et al.HIV clinic-based extended-release naltrexone versus treatment as usual for people with HIV and opioid use disorder: a non-blinded, randomized non-inferiority trial

    Korthuis, P. T.et al.HIV clinic-based extended-release naltrexone versus treatment as usual for people with HIV and opioid use disorder: a non-blinded, randomized non-inferiority trial. en.Addiction 117, 1961–1971. issn: 0965-2140, 1360-0443. https : / / onlinelibrary . wiley ...

    2023

  16. [24]

    V.et al.Sublingual Buprenorphine-Naloxone Compared With Injection Naltrexone for Opioid Use Disorder: Potential Utility of Patient Characteristics in Guiding Choice of Treatment

    Nunes, E. V.et al.Sublingual Buprenorphine-Naloxone Compared With Injection Naltrexone for Opioid Use Disorder: Potential Utility of Patient Characteristics in Guiding Choice of Treatment. en.American Journal of Psychiatry 178, 660–671. issn: 0002-953X, 1535-7228. http://ajp.p...

    2023

  17. [25]

    Martinez, R. A. M.et al.Conceptualization, Operationalization, and Utilization of Race and Ethnicity in Major Epidemiology Journals, 1995–2018: A Systematic Review. en.American Journal of Epidemi- ology 192, 483–496. issn: 0002-9262, 1476-6256. https://academic.oup.com/aje/art...

    2023

  18. [26]

    Jones, C. P. Invited commentary: "Race," racism, and the practice of epidemiology.American Journal of Epidemiology154, 299–304. issn: 00029262 (2001)

    2001

  19. [27]

    J., Polley, E

    Van Der Laan, M. J., Polley, E. C. & Hubbard, A. E. Super Learner. en.Statistical Applications in Genetics and Molecular Biology6, Article 25. issn: 21946302, 15446115 (2007)

    2007

  20. [28]

    mlr3superlearner: Super Learner Fitting and Prediction2024

    Williams, N. mlr3superlearner: Super Learner Fitting and Prediction2024

  21. [29]

    V., Van Der Laan, M

    Phillips, R. V., Van Der Laan, M. J., Lee, H. & Gruber, S. Practical considerations for specifying a super learner. en.International Journal of Epidemiology52, 1276–1285. issn: 0300-5771, 1464-3685. https://academic.oup.com/ije/article/52/4/1276/7076266 (2023) (Aug. 2023)

    2023

  22. [30]

    Dahabreh, I. J. et al. Sensitivity analysis using bias functions for studies extending inferences from a randomized trial to a target population. en.Statistics in Medicine42, 2029–2043. issn: 0277-6715, 1097-0258. https://onlinelibrary.wiley.com/doi/10.1002/sim.9550 (2024) (June 2023)

    2024 doi

  23. [31]

    Sensitivity Analysis in the Generalization of Experimental Resultsen

    Huang, M. Sensitivity Analysis in the Generalization of Experimental Resultsen. arXiv:2202.03408 [stat]. Feb. 2022.http://arxiv.org/abs/2202.03408 (2024)

    2024 arXiv

  24. [32]

    Q., Ebnesajjad, C., Cole, S

    Nguyen, T. Q., Ebnesajjad, C., Cole, S. R. & Stuart, E. A. Sensitivity analysis for an unobserved moderator in RCT-to-target-population generalization of treatment effects. en.The Annals of Applied Statistics 11. issn: 1932-6157. https : / / projecteuclid . org / journals / an...

    2024 doi

  25. [33]

    Q., Ackerman, B., Schmid, I., Cole, S

    Nguyen, T. Q., Ackerman, B., Schmid, I., Cole, S. R. & Stuart, E. A. Sensitivity analyses for effect modifiers not observed in the target population when generalizing treatment effects from a randomized controlled trial: Assumptions, models, effect scales, data scenarios, and ...

    2024 doi

  26. [34]

    H., Bodnar, L

    Zeng, Z., Kennedy, E. H., Bodnar, L. M. & Naimi, A. I.Efficient Generalization and Transportation en. arXiv:2302.00092 [stat]. Mar. 2023.http://arxiv.org/abs/2302.00092 (2024). 17 Tables Table 1.Characteristics of TEDS-A cohort and X:BOT sample Sample, N(%) TEDS-A X:BOT Charac...

    2024 arXiv

  27. [36]

    A3* =⇒ Pr(Y a = z, Za = y|W, S= 1) = Pr(Y a = y, Za = z|W, A= a, S= 1)

  28. [37]

    A3* =⇒ Pr (Z a = z|W, S= 1) = Pr (Z a = z|W, A= a, S= 1) X y Pr(Y a = y, Za = z|W, S= 1) = X y Pr(Y a = y, Za = z|W, A= a, S= 1) Pr(Z a = z|W, S= 1) = Pr(Z a = z|W, A= a, S= 1)

  29. [38]

    A3* =⇒ E(Y a|W, Za, S= 1) =E(Y a|W, A= a, Za, S= 1) Pr(Y a = y|W, Za = z, S= 1) =Pr(Y a = y, Za = z|W, S= 1) Pr(Z a = z|W, S= 1) = Pr(Y a = y, Za = z|W, A= a, S= 1) Pr(Z a = z|W, A= a, S= 1) by A3* = Pr(Y a = y|W, A= a, Za = z, S= 1)

  30. [39]

    First order error

    A3* =⇒ A3 X z Pr(Y a = y, Za = z|W, S= 1) = X z Pr(Y a = y, Za = z|W, A= a, S= 1) Pr(Y a = y|W, S= 1) = Pr(Y a = y|W, A= a, S= 1) Assumption A4* implies A4 A4*: Pr(A = a|W = w, Za = z, S= 1)> 0 when Pr(S = 0|W = w, Za = z) > 0 A4: Pr(A = a|W = w, S= 1)> 0 when Pr(S = 0|W = w) ...

    2014

  31. [2014]

    https : / / onlinelibrary

    issn: 0277-6715, 1097-0258. https : / / onlinelibrary . wiley . com / doi / 10 . 1002 / sim . 8426 (2024) (June 2020)

    2024

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.