arxiv: 2604.25064 · v1 · submitted 2026-04-27 · 📊 stat.ME

Evolving Longitudinal Patient Histories and Re-enrollment in Master Protocol Trials

Shiyu Wan , Yuhan Qian , Yanyao Yi , Nicole Mayer-Hamblett , Patrick J. Heagerty , Ting Ye This is my paper

Pith reviewed 2026-05-08 01:53 UTC · model grok-4.3

classification 📊 stat.ME

keywords master protocol trialsre-enrollmentestimandsrandomized trialslongitudinal datacausal inferenceadded-effectcystic fibrosis

0 comments p. Extension

The pith

Master protocol trials allowing re-enrollment can maintain randomized integrity by using an episode-specific definition of the eligible population.

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

This paper develops statistical methods for master protocol trials in which patients can participate in multiple episodes by re-enrolling for different therapies. The central innovation is defining the episode-specific entire concurrently eligible population to specify the target group for each episode in a way that keeps randomized comparisons intact. This population definition does not depend on how treatments are randomly assigned or the trial's operational setup. An added-effect estimand then sums up the effects from separate episodes to give an overall treatment benefit measure. Inference uses estimators that adjust for efficiency while properly handling correlations from the same patients appearing in multiple episodes, all under standard randomized trial assumptions.

Core claim

We define the episode-specific entire concurrently eligible (ECE) population, which preserves the integrity of randomized comparisons and remains invariant to randomization ratios and operational formats. We then introduce a per-episode added-effect estimand that aggregates episode-specific effects into an interpretable overall measure. For inference, we develop weighting and post-stratification estimators under the same minimal assumptions as conventional randomized trials, with model-assisted covariate adjustment to improve efficiency. We establish asymptotic distributions for all estimators and provide cluster-robust variance estimators that properly account for within-participant tration

What carries the argument

The episode-specific entire concurrently eligible (ECE) population that ensures randomized comparisons stay valid across episodes regardless of assignment mechanisms.

Load-bearing premise

Re-enrollment decisions introduce no unmeasured confounding with the outcomes beyond what covariate adjustments can handle.

What would settle it

A dataset or simulation in which re-enrollment is driven by unmeasured patient characteristics that also influence treatment response, leading to discrepant results between the proposed estimators and a first-episode-only analysis.

Figures

Figures reproduced from arXiv: 2604.25064 by Nicole Mayer-Hamblett, Patrick J. Heagerty, Shiyu Wan, Ting Ye, Yanyao Yi, Yuhan Qian.

**Figure 1.** Figure 1: Illustration of SIMPLIFY treatment-arm enrollment and follow-up. Participants are first randomized to a view at source ↗

**Figure 2.** Figure 2: Illustration of the SIMPLIFY DAG. The red path highlights the collider path opened by conditioning on view at source ↗

**Figure 3.** Figure 3: Estimates and 95% confidence intervals (in %) in the SIMPLIFY trial. view at source ↗

read the original abstract

A master protocol trial uses a single overarching protocol to test multiple therapies, often across several diseases or subtypes. Although such trials offer considerable flexibility and efficiency, their constrained and non-uniform treatment assignment raises two core challenges: precisely defining treatment effects and conducting robust, efficient inference. These challenges intensify when participants can re-enroll to receive additional eligible therapies over time. To address these issues, we first define a clinically meaningful estimand with a clear population specification for master protocol trials that allow re-enrollment across multiple episodes. Specifically, we define the episode-specific entire concurrently eligible (ECE) population, which preserves the integrity of randomized comparisons and remains invariant to randomization ratios and operational formats. We then introduce a per-episode added-effect estimand that aggregates episode-specific effects into an interpretable overall measure. For inference, we develop weighting and post-stratification estimators under the same minimal assumptions as conventional randomized trials, with model-assisted covariate adjustment to improve efficiency. We establish asymptotic distributions for all estimators and provide cluster-robust variance estimators that properly account for within-participant correlation induced by re-enrollment. We evaluate our methods through extensive simulations and apply our methods to SIMPLIFY, a master protocol trial comparing continuation versus discontinuation of two common cystic fibrosis therapies. All analyses are conducted using the \textsf{R} package \textsf{RobinCID}.

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.

Desk Editor's Note private letter to a colleague

The paper defines new estimands for re-enrollment in master protocols that maintain randomization integrity under standard assumptions.

read the letter

The main takeaway is that the authors introduce the episode-specific entire concurrently eligible population, which keeps randomized comparisons valid and invariant to assignment ratios, plus a per-episode added-effect estimand that combines effects across episodes for an overall measure in re-enrolling patients. What stands out is the development of weighting and post-stratification estimators under minimal assumptions from conventional trials, with model-assisted covariate adjustment. They establish asymptotic normality and provide cluster-robust variance estimators that account for within-participant correlation due to re-enrollment. The extensive simulations and the application to the SIMPLIFY master protocol trial in cystic fibrosis offer validation, and the RobinCID R package supports practical use. Soft spots are minor at most. Everything builds on conditional randomization and no unmeasured confounding, which the paper does not overclaim. Re-enrollment might raise practical issues with eligibility changes, but the methods address this through the defined population without circularity or hidden dependencies. This work is for clinical trial statisticians handling complex designs with multiple therapies and patient re-enrollment, particularly in oncology and rare disease research. A reader interested in estimands and inference for longitudinal trial data would get direct value from the new constructs and tools. The clear thinking and formal results mean it deserves serious peer review. I would recommend accepting it for peer review.

Referee Report

2 major / 4 minor

Summary. The paper addresses inference challenges in master protocol trials permitting re-enrollment by defining the episode-specific entire concurrently eligible (ECE) population, which preserves randomized comparisons and is invariant to assignment ratios and trial formats. It introduces a per-episode added-effect estimand that aggregates episode-specific effects. Weighting and post-stratification estimators are developed with model-assisted covariate adjustment under standard conditional randomization assumptions; asymptotic normality is established and cluster-robust variance estimators are provided to account for within-participant dependence induced by re-enrollment. Methods are validated via simulations and applied to the SIMPLIFY cystic fibrosis trial, with implementation in the R package RobinCID.

Significance. If the central constructions hold, the work offers a principled, clinically interpretable framework for treatment-effect estimation in re-enrolling master protocols while retaining the integrity of randomization. Strengths include reliance on minimal RCT assumptions, explicit handling of within-participant correlation via cluster-robust variances, extensive simulation validation, and provision of open-source software (RobinCID) for reproducibility. This could improve the design and analysis of flexible multi-therapy trials.

major comments (2)

[§3.2] §3.2, around the cluster-robust variance formula: the derivation assumes that re-enrollment decisions are independent of potential outcomes conditional on observed covariates and eligibility; however, the paper does not provide a formal sensitivity analysis or bounding argument for mild violations of this no-unmeasured-confounding condition, which is load-bearing for the claimed robustness of the variance estimator under realistic re-enrollment behavior.
[Table 4] Table 4 (SIMPLIFY application): the reported standard errors for the per-episode added-effect estimand appear to treat episodes as independent across participants, yet the cluster-robust adjustment is described only at the participant level; it is unclear whether the reported intervals fully incorporate the multi-episode structure within participants, which directly affects the inference claims for the overall measure.

minor comments (4)

[Abstract] Abstract: the phrase 'textsf{R} package' should be rendered consistently with the LaTeX formatting used elsewhere in the manuscript.
[§2.1] §2.1: the definition of the ECE population indicator would benefit from an explicit mathematical expression (e.g., an indicator function over concurrent eligibility) to make the invariance claim easier to verify.
[Figure 2] Figure 2: axis labels and legend entries for the different estimators (weighting vs. post-stratification) are difficult to distinguish at the printed size; consider adding a direct comparison panel.
[References] References: several citations to master-protocol methodology papers are missing page numbers or DOIs, which should be completed for consistency.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments on our manuscript. We address each major comment below, indicating the changes we will make to the revised version.

read point-by-point responses

Referee: [§3.2] §3.2, around the cluster-robust variance formula: the derivation assumes that re-enrollment decisions are independent of potential outcomes conditional on observed covariates and eligibility; however, the paper does not provide a formal sensitivity analysis or bounding argument for mild violations of this no-unmeasured-confounding condition, which is load-bearing for the claimed robustness of the variance estimator under realistic re-enrollment behavior.

Authors: We thank the referee for highlighting this assumption. The cluster-robust variance estimator is derived under the conditional independence of re-enrollment decisions from potential outcomes given observed covariates and eligibility, which aligns with the minimal assumptions stated for the randomized trial framework. We agree that the manuscript would benefit from explicit acknowledgment of this point. In the revision, we will add a paragraph in Section 3.2 stating the assumption clearly and noting that mild violations could be investigated via simulation-based sensitivity analyses or bounding methods in future work. This addition will better delineate the scope of the robustness claims. revision: partial
Referee: [Table 4] Table 4 (SIMPLIFY application): the reported standard errors for the per-episode added-effect estimand appear to treat episodes as independent across participants, yet the cluster-robust adjustment is described only at the participant level; it is unclear whether the reported intervals fully incorporate the multi-episode structure within participants, which directly affects the inference claims for the overall measure.

Authors: We appreciate the referee's attention to the variance implementation in the application. The standard errors in Table 4 are obtained from the participant-level cluster-robust variance estimator described in Section 3.2, which explicitly accounts for within-participant dependence across multiple episodes. This is implemented via the sandwich estimator in the RobinCID package and used for all reported results. To remove any ambiguity, we will update the text in Section 5 and the caption of Table 4 to state explicitly that variances are computed with participant-level clustering. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation is self-contained

full rationale

The paper defines the ECE population and per-episode added-effect estimand directly from the master protocol structure, randomization, and re-enrollment mechanics. These are not fitted to data or defined in terms of the estimators. Inference uses standard weighting/post-stratification with covariate adjustment and cluster-robust variances, all justified under the same conditional randomization and no-unmeasured-confounding assumptions as ordinary RCTs. Asymptotics follow from established statistical theory for dependent data rather than any self-referential construction. Simulations and the SIMPLIFY application provide external validation. No load-bearing step reduces claimed results to inputs by definition, self-citation, or fitted-parameter renaming.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 2 invented entities

The central claims rest on standard randomization assumptions from conventional trials plus new definitional constructs for the ECE population and added-effect estimand. No free parameters are explicitly fitted in the abstract description; the estimators use weighting and post-stratification under minimal assumptions.

axioms (2)

domain assumption Treatment assignment is random conditional on eligibility at each episode
Invoked to preserve integrity of randomized comparisons in the ECE population definition.
domain assumption Re-enrollment decisions do not introduce unmeasured confounding beyond covariate adjustment
Required for the weighting and post-stratification estimators to be valid.

invented entities (2)

Episode-specific entire concurrently eligible (ECE) population no independent evidence
purpose: To define a population that preserves randomized comparisons invariant to randomization ratios
New definitional construct introduced to handle re-enrollment without bias from changing eligibility.
Per-episode added-effect estimand no independent evidence
purpose: To aggregate episode-specific effects into an overall interpretable measure
New estimand to summarize treatment effects across multiple episodes.

pith-pipeline@v0.9.0 · 5555 in / 1667 out tokens · 36623 ms · 2026-05-08T01:53:43.998489+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

24 extracted references · 21 canonical work pages

[1]

Janet Woodcock and Lisa M LaVange

doi: 10.1001/jama.2015.2316. Janet Woodcock and Lisa M LaVange. Master protocols to study multiple therapies, multiple diseases, or both.New England Journal of Medicine, 377(1):62–70,

work page doi:10.1001/jama.2015.2316 2015
[2]

doi: 10.1056/NEJMra1510062. U.S. Food and Drug Administration. Master protocols for drug and biological product development. FDA Guidance Document, December

work page doi:10.1056/nejmra1510062
[3]

Sawsan Rashdan and David E Gerber

doi: 10.1093/annonc/ mdw413. Sawsan Rashdan and David E Gerber. Going into battle: umbrella and basket clinical trials to accelerate the study of biomarker-based therapies.Annals of Translational Medicine, 4(24):529, dec

work page doi:10.1093/annonc/
[4]

doi: 10.21037/atm.2016.12.57. Brian P. Hobbs, Renata C. Pestana, Emily C. Zabor, Alexander M. Kaizer, and David S. Hong. Basket trials: Review of current practice and innovations for future trials.Journal of Clinical Oncology, 40(30):3520–3528,

work page doi:10.21037/atm.2016.12.57 2016
[5]

Amanda J

doi: 10.1200/JCO.21.02285. Amanda J. Redig and Pasi A. Jänne. Basket trials and the evolution of clinical trial design in an era of genomic medicine.Journal of Clinical Oncology, 33(9):975–977,

work page doi:10.1200/jco.21.02285
[6]

URL https: //ascopubs.org/doi/abs/10.1200/JCO.2014.59.8433

doi: 10.1200/JCO.2014.59.8433. URL https: //ascopubs.org/doi/abs/10.1200/JCO.2014.59.8433. PMID: 25667288. Benjamin R. Saville and Scott M. Berry. Efficiencies of platform clinical trials: A vision of the future.Clinical Trials, 13(3):358–366,

work page doi:10.1200/jco.2014.59.8433 2014
[7]

doi: 10.1177/1740774515626362. J. J. H. Park, E. Siden, M. J. Zoratti, et al. Systematic review of basket trials, umbrella trials, and platform trials: a landscape analysis of master protocols.Trials, 20:572,

work page doi:10.1177/1740774515626362
[8]

Adrian Wells, Peter Fisher, Sarah Myers, Jane Wheatley, Tamsin Patel, and Chris R

doi: 10.1186/s13063-019-3664-1. Adrian Wells, Peter Fisher, Sarah Myers, Jane Wheatley, Tamsin Patel, and Chris R. Brewin. Metacognitive therapy in treatment-resistant depression: a platform trial.Behaviour Research and Therapy, 50(6):367–373,

work page doi:10.1186/s13063-019-3664-1
[9]

URLhttps://doi.org/10.1016/j.brat.2012.02.004

doi: 10.1016/j.brat.2012.02.004. URLhttps://doi.org/10.1016/j.brat.2012.02.004. Sharon-Lise T. Normand. The recovery platform.New England Journal of Medicine, 384(8):757–758,

work page doi:10.1016/j.brat.2012.02.004 2012
[10]

URLhttps://www.nejm.org/doi/full/10.1056/NEJMe2025674

doi: 10.1056/NEJMe2025674. URLhttps://www.nejm.org/doi/full/10.1056/NEJMe2025674. Stefan M. Gold, Marta Bofill Roig, J. Jaime Miranda, Carmine Pariante, Martin Posch, and Christian Otte. Platform trials and the future of evaluating therapeutic behavioural interventions.Nature reviews psychology, 1(1):7–8,

work page doi:10.1056/nejme2025674
[11]

URL https://doi.org/10.1093/brain/awad063

doi: 10.1093/brain/awad063. URL https://doi.org/10.1093/brain/awad063. Sailish Honap, Bruce E Sands, Vipul Jairath, Silvio Danese, Eric Vicaut, and Laurent Peyrin-Biroulet. Basket, umbrella, and platform trials: The potential for master protocol–based trials in inflammatory bowel disease.Gastroenterology, 167(4):636–642,

work page doi:10.1093/brain/awad063
[12]

URL https: //doi.org/10.1016/S1470-2045(20)30475-7

doi: 10.1016/S1470-2045(20)30475-7. URL https: //doi.org/10.1016/S1470-2045(20)30475-7. Nicole Mayer-Hamblett, David P. Nichols, Katherine Odem-Davis, et al. Evaluating the impact of stopping chronic therapies after modulator drug therapy in cystic fibrosis: The SIMPLIFY clinical trial study design.Annals of the American Thoracic Society, 18(8):1397–1405,

work page doi:10.1016/s1470-2045(20)30475-7 2045
[13]

URL https://doi.org/10.1513/AnnalsATS.202010-1336SD

doi: 10.1513/AnnalsATS.202010-1336SD. URL https://doi.org/10.1513/AnnalsATS.202010-1336SD. L. O. Ouma, J. M. S. Wason, H. Zheng, N. Wilson, and M. Grayling. Design and analysis of umbrella trials: Where do we stand?Frontiers in Medicine, 9:1037439,

work page doi:10.1513/annalsats.202010-1336sd
[14]

URL https: //doi.org/10.3389/fmed.2022.1037439

doi: 10.3389/fmed.2022.1037439. URL https: //doi.org/10.3389/fmed.2022.1037439. Yuhan Qian, Yifan Yi, Jun Shao, Yanyao Yi, Gregory Levin, Nicole Mayer-Hamblett, Patrick J. Heagerty, and Ting Ye. From estimands to robust inference of treatment effects in platform trials.arXiv preprint arXiv:2411.12944,

work page doi:10.3389/fmed.2022.1037439 2022
[15]

doi: 10.1177/17407745221112013

ISSN 1740-7745. doi: 10.1177/17407745221112013. URLhttps://doi.org/10.1177/17407745221112013. Marta Bofill Roig, Cora Burgwinkel, Ursula Garczarek, Franz Koenig, Martin Posch, Quynh Nguyen, and Katharina Hees. On the use of non-concurrent controls in platform trials: a scoping review.Trials, 24(1):408,

work page doi:10.1177/17407745221112013
[16]

doi: 10.1111/biom.13680

ISSN 1541-0420. doi: 10.1111/biom.13680. URLhttps://onlinelibrary.wiley.com/doi/abs/10.1111/biom.13680. Jingjing Liu, Cong Lu, Zhihua Jiang, Daniel Alemayehu, Lei Nie, and Haitao Chu. Borrowing concurrent information from non-concurrent control to enhance statistical efficiency in platform trials.Current Oncology, 30(4):3964–3973,

work page doi:10.1111/biom.13680
[17]

URLhttps://doi.org/10.3390/curroncol30040300

doi: 10.3390/curroncol30040300. URLhttps://doi.org/10.3390/curroncol30040300. Beibei Guo, Li Wang, and Ying Yuan. Treatment comparisons in adaptive platform trials adjusting for temporal drift. Statistics in Biopharmaceutical Research, pages 1–10,

work page doi:10.3390/curroncol30040300
[18]

URLhttps://onlinelibrary.wiley.com/doi/abs/10.1002/sim.70017

doi: https: //doi.org/10.1002/sim.70017. URLhttps://onlinelibrary.wiley.com/doi/abs/10.1002/sim.70017. N Mayer-Hamblett, F Ratjen, R Russell, et al. Discontinuation versus continuation of hypertonic saline or dornase alfa in modulator-treated people with cystic fibrosis (simplify): results from two parallel, multicentre, open-label, randomised, controlled...

work page doi:10.1002/sim.70017
[19]

URLhttps://doi.org/10.1016/S2213-2600(22)00434-9

doi: 10.1016/ S2213-2600(22)00434-9. URLhttps://doi.org/10.1016/S2213-2600(22)00434-9. ICH. E9 statistical principles for clinical trials. Guidance for Industry, 09

work page doi:10.1016/s2213-2600(22)00434-9
[20]

Brennan C

ICH E9 guidance (originally issued September 16, 1998). Brennan C. Kahan, Ian R. White, Richard Hooper, and Sandra Eldridge. Re-randomisation trials in multi-episode settings: Estimands and independence estimators.Statistical Methods in Medical Research, 31(7):1342–1354,

1998
[21]

an essay on the logical foundations of survey sampling, part one

doi: 10.1177/09622802221094140. URLhttps://doi.org/10.1177/09622802221094140. Jaroslav Hájek. Comment on “an essay on the logical foundations of survey sampling, part one”. InThe Foundations of Survey Sampling, page

work page doi:10.1177/09622802221094140
[22]

URLhttps://doi.org/10.1080/01621459.2022.2049278

doi: 10.1080/01621459.2022.2049278. URLhttps://doi.org/10.1080/01621459.2022.2049278. Marlena Bannick, Yuhan Qian, Ting Ye, Yanyao Yi, and Faith Bian.RobinCar: Robust Inference for Covariate Adjustment in Randomized Clinical Trials,

work page doi:10.1080/01621459.2022.2049278 2022
[23]

I(G i,t =h)·E(Y i,t|Ai,t =j, G i,t =h) #   = TX t=1 E    HjktX h=1

If ˆµjk is obtained from a finite-dimensional parametric model or satisfies the Donsker condition, then gn belongs to a P0-Donsker class,. Hence, by Lemma 19.24 in van der Vaart (1998), we have √n ˜∆jk =o p(1). Consequently, √n∆jk =o p(1) and √n[ˆθaipw jk −θ jk] converge in distribution to N(0,Var[ϕaipw jk (Ri)]). Proof of Theorem 1 d) Theorem 1 d):Asympt...

1998
[24]

Moreover, in the proof of Theorem 1 c) we showed that √n( 1 njk /n − 1PT t=1 P(Ijkt(i)=1)) =O p(1)

× 1 n nX i=1    TX t=1 Ijkt(i)    1− HjktX h=1 I(G i,t =h)·I(A i,t =j) ˆP(Ai,t =j|G i,t =h)   ·[ˆµjkt(XXX i,t)−µ jkt(XXX i,t)]      , 29 APREPRINT- By the Cauchy–Schwarz inequality, we obtain: 1 n nX i=1    TX t=1 Ijkt(i)    1− HjktX h=1 I(G i,t =h, A i,t =j) ˆP(Ai,t =j|G i,t =h)   ·[ˆµjk(XXX i,t)−µ jk(XXX i,t)]      ≤    ...

1998