arxiv: 2605.14222 · v1 · submitted 2026-05-14 · 📊 stat.ME

Recognition: 2 theorem links

· Lean Theorem

Robust and Data-Adaptive Integration of Nonconcurrent Data in Platform Trials via Gaussian Processes

Yuhan Qian , Yu Du , Jingning Zhang , Yanyao Yi , Patrick J. Heagerty , Ting Ye

Authors on Pith no claims yet

Pith reviewed 2026-05-15 02:43 UTC · model grok-4.3

classification 📊 stat.ME

keywords platform trialsGaussian processesnonconcurrent datatreatment effectbias controlclinical trial designdata integration

0 comments

The pith

Gaussian processes enable data-adaptive integration of nonconcurrent controls in platform trials while bounding bias.

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

Platform trials evaluate multiple treatments continuously but often exclude nonconcurrent data to avoid bias from time trends. This paper proposes a Gaussian process framework that models smooth temporal trends to incorporate such data adaptively. It provides uncertainty quantification and a frequentist interpretation via kernel ridge regression. Theoretical guarantees show reduced posterior variance and controlled bias. The approach is demonstrated on a hypothetical trial and implemented in an R package.

Core claim

The framework uses single-task and multi-task Gaussian processes to integrate nonconcurrent control data, exploiting temporal smoothness for data-adaptive borrowing. This yields lower variance in treatment effect estimates with bias bounded by a non-increasing function of the time gap.

What carries the argument

Gaussian process prior on the temporal trend function, which induces a kernel that weights nonconcurrent observations based on their similarity in time.

If this is right

Adding nonconcurrent controls reduces the posterior variance of the treatment effect estimate.
The bias introduced is controlled by a bound that does not increase with the time separation.
The method extends naturally to discrete outcomes and covariate-adjusted analyses.
Implementation allows practical application in ongoing platform trials.

Where Pith is reading between the lines

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

Trials could potentially enroll fewer concurrent participants by leveraging historical controls within the platform.
This integration strategy might apply to other master protocol designs with staggered entry.
Further work could explore robustness to violations of smoothness assumptions through sensitivity analyses.

Load-bearing premise

Temporal trends in patient outcomes must be smooth enough that a Gaussian process can model them without leaving substantial unaccounted bias from nonconcurrent periods.

What would settle it

Simulations or real data where nonconcurrent data addition increases the mean squared error beyond what the bound predicts, or where variance does not decrease as expected under smooth trends.

Figures

Figures reproduced from arXiv: 2605.14222 by Jingning Zhang, Patrick J. Heagerty, Ting Ye, Yanyao Yi, Yu Du, Yuhan Qian.

**Figure 2.** Figure 2: Estimation results under one simulated dataset from the reference setting. [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗

read the original abstract

A platform trial is an innovative clinical trial design that enables simultaneous and continuous evaluation of multiple treatments within a single master protocol. Existing robust methods restrict analyses to concurrently randomized participants due to concerns that including nonconcurrent data may introduce bias from temporal trends. However, this exclusion represents a missed opportunity to improve efficiency. We propose a Gaussian process framework for incorporating nonconcurrent data that exploits temporal smoothness, a key feature of platform trials. The framework includes single-task and multi-task formulations and provides data-adaptive integration of nonconcurrent data with uncertainty quantification. The connection to kernel ridge regression yields a transparent frequentist interpretation of how nonconcurrent data are integrated. We establish two theoretical guarantees: incorporating nonconcurrent controls reduces the posterior variance of the treatment effect, and the resulting bias is controlled by a non-increasing bound. We extend the framework to discrete outcomes and to covariate adjustment, illustrate it on a hypothetical platform trial constructed from SURMOUNT-1, and provide an implementation in the R package 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

GP framework for nonconcurrent data in platform trials gives usable variance reduction under the smoothness assumption, with an R package attached.

read the letter

The core takeaway is that this paper shows how to pull nonconcurrent controls into platform trial analyses via Gaussian processes without the usual hard exclusion. It claims posterior variance drops while bias stays bounded by a non-increasing function, using both single-task and multi-task formulations plus a kernel ridge regression link for frequentist clarity. They also extend it to discrete outcomes and covariate adjustment, supply an R package called RobinCID, and run an illustration on a constructed SURMOUNT-1 example. That package and the worked example are the practical wins here; trial statisticians can actually try it out. The theoretical guarantees follow from standard GP posterior variance and bias results, which is clean when the kernel matches the temporal trend. The data-adaptive weighting is a step beyond methods that simply drop nonconcurrent periods. The soft spot is exactly the one the stress-test note flags: everything rests on the true temporal function being well approximated by the chosen kernel's RKHS. If the trend has jumps or higher-frequency variation the kernel misses, the bias bound can stop being non-increasing and you trade variance for uncontrolled bias. The paper does not appear to include quantitative sensitivity checks or worst-case bounds under misspecification, so the guarantees read as conditional on correct specification. Simulations likely assume the model holds, which is common but leaves real-data performance open. This is aimed at methodologists and statisticians who run or analyze platform trials and want to use all available controls. A reader already comfortable with GPs will see the value quickly. It deserves serious peer review because the idea is targeted, the implementation is public, and the theory is stated explicitly, even if the kernel assumption needs more stress-testing in revision.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes a Gaussian process framework, with single-task and multi-task formulations, for data-adaptive integration of nonconcurrent control data in platform trials by exploiting temporal smoothness. It connects the approach to kernel ridge regression for a frequentist interpretation, establishes two theoretical guarantees (posterior variance reduction for the treatment effect and a non-increasing bias bound), extends the method to discrete outcomes and covariate adjustment, illustrates it on a hypothetical platform trial derived from SURMOUNT-1, and provides an R package implementation (RobinCID).

Significance. If the theoretical guarantees hold under the smoothness assumption, the framework could meaningfully improve efficiency in platform trials by allowing principled use of nonconcurrent data with uncertainty quantification, potentially reducing required sample sizes while maintaining robustness. The explicit link to kernel ridge regression and the reproducible software are notable strengths that support practical adoption.

major comments (2)

[§3] §3 (theoretical guarantees): The non-increasing bias bound and variance reduction claims rest on the assumption that the unknown temporal trend lies in (or is well-approximated by) the RKHS of the chosen kernel. The manuscript does not provide a quantitative sensitivity analysis or worst-case bound under kernel misspecification (e.g., abrupt changes or higher-frequency variation), which is load-bearing for the central claim that bias remains controlled.
[§5] §5 (simulation and illustration): The data exclusion rules for nonconcurrent periods, the procedure for selecting or estimating GP kernel hyperparameters, and the exact construction of the hypothetical SURMOUNT-1 platform trial are not fully specified. This prevents independent verification of the reported performance gains and the practical behavior of the data-adaptive integration.

minor comments (2)

[Abstract] The abstract states that the bias is 'controlled by a non-increasing bound' but does not name the theorem or section containing the formal statement.
[§2] Notation distinguishing the single-task and multi-task GP models (e.g., covariance functions and posterior expressions) would benefit from an explicit comparison table early in the methods section.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the thorough review and positive recommendation regarding the significance of our work. We have carefully considered the major comments and will make revisions to address them, as detailed in the point-by-point responses below. These changes will strengthen the theoretical robustness and reproducibility of the manuscript.

read point-by-point responses

Referee: [§3] §3 (theoretical guarantees): The non-increasing bias bound and variance reduction claims rest on the assumption that the unknown temporal trend lies in (or is well-approximated by) the RKHS of the chosen kernel. The manuscript does not provide a quantitative sensitivity analysis or worst-case bound under kernel misspecification (e.g., abrupt changes or higher-frequency variation), which is load-bearing for the central claim that bias remains controlled.

Authors: We thank the referee for highlighting this important point. The theoretical guarantees in §3 are indeed derived under the assumption that the temporal trend belongs to the reproducing kernel Hilbert space (RKHS) induced by the chosen kernel, which encodes the smoothness assumption. This is a standard assumption in Gaussian process regression and kernel methods, and we believe it is plausible in the context of platform trials where temporal trends are typically smooth due to gradual changes in patient populations or standards of care. However, to address concerns about kernel misspecification, we will add a new subsection in the revised manuscript that includes a quantitative sensitivity analysis. This will involve simulations where the true trend deviates from the RKHS (e.g., with abrupt changes or higher-frequency components) and evaluate the resulting bias and variance. We will also discuss the choice of kernel and potential robustness measures, such as using more flexible kernels or cross-validation for kernel selection. revision: yes
Referee: [§5] §5 (simulation and illustration): The data exclusion rules for nonconcurrent periods, the procedure for selecting or estimating GP kernel hyperparameters, and the exact construction of the hypothetical SURMOUNT-1 platform trial are not fully specified. This prevents independent verification of the reported performance gains and the practical behavior of the data-adaptive integration.

Authors: We agree that additional details are necessary for full reproducibility. In the revised manuscript, we will expand §5 and add a dedicated appendix that specifies: (1) the exact data exclusion rules for nonconcurrent periods, including any criteria based on time windows or patient characteristics; (2) the procedure for selecting or estimating GP kernel hyperparameters, which involves maximizing the marginal likelihood with details on optimization and initialization; and (3) the precise construction of the hypothetical platform trial derived from the SURMOUNT-1 dataset, including how nonconcurrent controls were simulated or selected, the temporal structure imposed, and any assumptions made. These additions will enable independent verification and clarify the practical implementation of the data-adaptive integration. revision: yes

Circularity Check

0 steps flagged

No circularity; theoretical guarantees derive directly from standard GP posterior properties

full rationale

The paper's central claims—posterior variance reduction when adding nonconcurrent controls and a non-increasing bias bound—are obtained from the explicit posterior formulas of the single-task and multi-task Gaussian process models. These derivations rely on the standard reproducing-kernel Hilbert space properties of the chosen kernel and the inclusion of additional observations; they do not reduce to fitted parameters renamed as predictions, self-definitional loops, or load-bearing self-citations. The connection to kernel ridge regression is presented as an interpretive equivalence rather than a circular justification. The smoothness assumption is stated upfront and the guarantees hold conditionally on that assumption, without smuggling the target result into the premise.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the domain assumption of temporal smoothness in trial data and standard Gaussian process modeling assumptions; kernel hyperparameters are fitted from data.

free parameters (1)

GP kernel hyperparameters
Length-scale, variance, and other kernel parameters are estimated from the data to achieve data-adaptive integration.

axioms (1)

domain assumption Temporal trends in platform trial outcomes are smooth enough to be modeled by a Gaussian process prior.
This smoothness enables the data-adaptive borrowing of nonconcurrent controls without uncontrolled bias.

pith-pipeline@v0.9.0 · 5486 in / 1331 out tokens · 32343 ms · 2026-05-15T02:43:51.057055+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

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

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

We establish two theoretical guarantees: incorporating nonconcurrent controls reduces the posterior variance of the treatment effect, and the resulting bias is controlled by a non-increasing bound.
IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

The connection to kernel ridge regression yields a transparent frequentist interpretation...

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

52 extracted references · 52 canonical work pages · 2 internal anchors

[1]

Kernels for Vector-Valued Functions: a Review

Alvarez, M. A., Rosasco, L., and Lawrence, N. D. (2012). Kernels for Vector - Valued Functions : a Review . arXiv:1106.6251 [stat]

work page internal anchor Pith review Pith/arXiv arXiv 2012
[2]

Aronszajn, N. (1950). Theory of reproducing kernels. Transactions of the American mathematical society 68, 337--404

work page 1950
[3]

S., Xi, D., Ye, T., and Yi, Y

Bannick, M., Bian, Y., Chen, G., Li, L., Qian, Y., Bové, D. S., Xi, D., Ye, T., and Yi, Y. (2026). The robincar family: R tools for robust covariate adjustment in randomized clinical trials

work page 2026
[4]

Barker, A., Sigman, C., Kelloff, G., Hylton, N., Berry, D., and Esserman, L. (2009). I-spy 2: an adaptive breast cancer trial design in the setting of neoadjuvant chemotherapy. Clinical Pharmacology & Therapeutics 86, 97--100

work page 2009
[5]

H., Tomashek, K

Beigel, J. H., Tomashek, K. M., Dodd, L. E., Mehta, A. K., Zingman, B. S., Kalil, A. C., Hohmann, E., Chu, H. Y., Luetkemeyer, A., Kline, S., et al. (2020). Remdesivir for the treatment of covid-19. New England Journal of Medicine 383, 1813--1826

work page 2020
[6]

M., Connor, J

Berry, S. M., Connor, J. T., and Lewis, R. J. (2015). The platform trial: an efficient strategy for evaluating multiple treatments. Journal of the American Medical Association 313, 1619--1620

work page 2015
[7]

Betancourt, M. (2020). Robust gaussian process modeling. https://betanalpha.github.io/assets/case_studies/gaussian_processes.html

work page 2020
[8]

Bofill Roig, M., Burgwinkel, C., Garczarek, U., Koenig, F., Posch, M., Nguyen, Q., and Hees, K. (2023). On the use of non-concurrent controls in platform trials: a scoping review. Trials 24, 408

work page 2023
[9]

Bofill Roig, M., K \"o nig, F., Meyer, E., and Posch, M. (2022). Commentary: Two approaches to analyze platform trials incorporating non-concurrent controls with a common assumption. Clinical Trials 19, 502--503

work page 2022
[10]

M., Hees, K., Jacko, P., Koenig, F., Magirr, D., Mesenbrink, P., et al

Bofill Roig, M., Krotka, P., Burman, C.-F., Glimm, E., Gold, S. M., Hees, K., Jacko, P., Koenig, F., Magirr, D., Mesenbrink, P., et al. (2022). On model-based time trend adjustments in platform trials with non-concurrent controls. BMC Medical Research Methodology 22, 1--16

work page 2022
[11]

Burki, T. (2023). Platform trials: the future of medical research? The Lancet Respiratory Medicine 11, 232--233

work page 2023
[12]

B \"u rkner, P.-C. (2017). brms: An r package for bayesian multilevel models using stan. Journal of statistical software 80, 1--28

work page 2017
[13]

E., Freidlin, B., and Korn, E

Dodd, L. E., Freidlin, B., and Korn, E. L. (2021). Platform trials—beware the noncomparable control group. New England Journal of Medicine 384, 1572--1573

work page 2021
[14]

Duvenaud, D. K. (2014). Automatic model construction with gaussian processes. Ph.D. thesis

work page 2014
[15]

Master protocols for drug and biological product development

FDA (2023). Master protocols for drug and biological product development. Draft Guidance for Industry. Center for Drug Evaluation and Research and Center for Biologics Evaluation and Research, Food and Drug Administration (FDA), U.S. Department of Health and Human Services. December 2023

work page 2023
[16]

Gao, C., Yang, S., Shan, M., Ye, W., Lipkovich, I., and Faries, D. (2025). Improving randomized controlled trial analysis via data-adaptive borrowing. Biometrika 112, asae069

work page 2025
[17]

M., Bofill Roig, M., Miranda, J

Gold, S. M., Bofill Roig, M., Miranda, J. J., Pariante, C., Posch, M., and Otte, C. (2022). Platform trials and the future of evaluating therapeutic behavioural interventions. Nature Reviews Psychology 1, 7--8

work page 2022
[18]

S., Gandara, D

Herbst, R. S., Gandara, D. R., Hirsch, F. R., Redman, M. W., LeBlanc, M., Mack, P. C., Schwartz, L. H., Vokes, E., Ramalingam, S. S., Bradley, J. D., et al. (2015). Lung master protocol (lung-map)—a biomarker-driven protocol for accelerating development of therapies for squamous cell lung cancer: Swog s1400. Clinical Cancer Research 21, 1514--1524

work page 2015
[19]

M., Aronne, L

Jastreboff, A. M., Aronne, L. J., Ahmad, N. N., Wharton, S., Connery, L., Alves, B., Kiyosue, A., Zhang, S., Liu, B., Bunck, M. C., and Stefanski, A. (2022). Tirzepatide once weekly for the treatment of obesity. New England Journal of Medicine 387, 205--216

work page 2022
[20]

Kanagawa, M., Hennig, P., Sejdinovic, D., and Sriperumbudur, B. K. (2018). Gaussian Processes and Kernel Methods : A Review on Connections and Equivalences . arXiv:1807.02582 [cs, stat]

work page internal anchor Pith review Pith/arXiv arXiv 2018
[21]

Kanagawa, M., Hennig, P., Sejdinovic, D., and Sriperumbudur, B. K. (2025). Gaussian processes and reproducing kernels: Connections and equivalences

work page 2025
[22]

Kish, L. (1965). Survey sampling. Wiley

work page 1965
[23]

Krotka, P., Bofill Roig , M., Hees, K., Jacko, P., and Magirr, D. (2024). NCC: Simulation and Analysis of Platform Trials with Non-Concurrent Controls . R package version 1.0, https://github.com/pavlakrotka/NCC

work page 2024
[24]

Lee, K. M. and Wason, J. (2020). Including non-concurrent control patients in the analysis of platform trials: is it worth it? BMC Medical Research Methodology 20, 165

work page 2020
[25]

Li, X., Miao, W., Lu, F., and Zhou, X.-H. (2023). Improving efficiency of inference in clinical trials with external control data. Biometrics 79, 394--403

work page 2023
[26]

Minka, T. P. (2001). A family of algorithms for approximate bayesian inference . PhD thesis, MIT, USA. AAI0803033

work page 2001
[27]

Neal, R. M. (1996). Bayesian Learning for Neural Networks . Springer-Verlag, Berlin, Heidelberg

work page 1996
[28]

Neal, R. M. (1999). Regression and classification using gaussian process priors. In Bayesian Statistics 6: Proceedings of the Sixth Valencia International Meeting June 6-10, 1998 . Oxford University Press

work page 1999
[29]

Nelder, J. A. and Wedderburn, R. W. M. (1972). Generalized linear models. Journal of the Royal Statistical Society. Series A (General) 135, 370--384

work page 1972
[30]

and Roy, J

Oganisian, A. and Roy, J. A. (2021). A practical introduction to Bayesian estimation of causal effects: Parametric and nonparametric approaches. Statistics in Medicine 40, 518--551

work page 2021
[31]

Pan, T., Mei, X., Zhang, X., Shen, W., and Ye, T. (2024). A bayesian approach for selecting relevant external data (base): Application to a study of long-term outcomes in a hemophilia gene therapy trial. arXiv preprint arXiv:2403.13260

work page arXiv 2024
[32]

J., Detry, M

Park, J. J., Detry, M. A., Murthy, S., Guyatt, G., and Mills, E. J. (2022). How to use and interpret the results of a platform trial: users’ guide to the medical literature. Jama 327, 67--74

work page 2022
[33]

Porcu, E., Bevilacqua, M., Schaback, R., and Oates, C. J. (2023). The mat\'ern model: A journey through statistics, numerical analysis and machine learning

work page 2023
[34]

J., and Ye, T

Qian, Y., Yi, Y., Shao, J., Yi, Y., Levin, G., Mayer-Hamblett, N., Heagerty, P. J., and Ye, T. (2025). From estimands to robust inference of treatment effects in platform trials

work page 2025
[35]

and Williams, C

Rasmussen, C. and Williams, C. (2005). Gaussian Processes for Machine Learning . Adaptive Computation and Machine Learning series. MIT Press

work page 2005
[36]

Dexamethasone in hospitalized patients with covid-19

RECOVERY Collaborative Group (2021). Dexamethasone in hospitalized patients with covid-19. New England journal of medicine 384, 693--704

work page 2021
[37]

R., Solin, A., and Vehtari, A

Riutort-Mayol, G., B \"u rkner, P.-C., Andersen, M. R., Solin, A., and Vehtari, A. (2023). Practical hilbert space approximate bayesian gaussian processes for probabilistic programming. Statistics and Computing 33, 17

work page 2023
[38]

B., Krotka, P., Hees, K., Koenig, F., Magirr, D., Jacko, P., Parke, T., and Posch, M

Roig, M. B., Krotka, P., Hees, K., Koenig, F., Magirr, D., Jacko, P., Parke, T., and Posch, M. (2025). Treatment-control comparisons in platform trials including non-concurrent controls. Statistics in Biopharmaceutical Research 0, 1--18

work page 2025
[39]

Rubin, D. B. (1974). Estimating causal effects of treatments in randomized and nonrandomized studies. Journal of Educational Psychology 66, 688

work page 1974
[40]

Rubin, D. B. (1981). The Bayesian Bootstrap . The Annals of Statistics 9, 130 -- 134

work page 1981
[41]

Rubin, D. B. (1987). Multiple Imputation for Nonresponse in Surveys . John Wiley & Sons Inc., New York

work page 1987
[42]

M., Zhang, X., and D \' az, I

Santacatterina, M., Giron, F. M., Zhang, X., and D \' az, I. (2025). Identification and estimation of causal effects using non-concurrent controls in platform trials. Statistics in Medicine 44, e70017

work page 2025
[43]

R., Berry, D

Saville, B. R., Berry, D. A., Berry, N. S., Viele, K., and Berry, S. M. (2022). The Bayesian Time Machine : Accounting for temporal drift in multi-arm platform trials. Clinical Trials 19, 490--501

work page 2022
[44]

Saville, B. R. and Berry, S. M. (2016). Efficiencies of platform clinical trials: a vision of the future. Clinical Trials 13, 358--366

work page 2016
[45]

A., Thomas, M., Cassidy, A., Weber, S., and Bretz, F

Schmidli, H., H \"a ring, D. A., Thomas, M., Cassidy, A., Weber, S., and Bretz, F. (2020). Beyond randomized clinical trials: use of external controls. Clinical Pharmacology & Therapeutics 107, 806--816

work page 2020
[46]

M., and Speed, T

Splawa-Neyman, J., Dabrowska, D. M., and Speed, T. P. (1990). On the application of probability theory to agricultural experiments. essay on principles. section 9. Statistical Science pages 465--472

work page 1990
[47]

L., Gwise, T., Hess, L., et al

Sridhara, R., Marchenko, O., Jiang, Q., Pazdur, R., Posch, M., Berry, S., Theoret, M., Shen, Y. L., Gwise, T., Hess, L., et al. (2022). Use of nonconcurrent common control in master protocols in oncology trials: report of an american statistical association biopharmaceutical section open forum discussion. Statistics in Biopharmaceutical Research 14, 353--357

work page 2022
[48]

L., and Soon, G

Wang, C., Lin, M., Rosner, G. L., and Soon, G. (2023). A bayesian model with application for adaptive platform trials having temporal changes. Biometrics 79, 1446--1458

work page 2023
[49]

and Barber, D

Williams, C. and Barber, D. (1998). Bayesian classification with gaussian processes. IEEE Transactions on Pattern Analysis and Machine Intelligence 20, 1342--1351

work page 1998
[50]

and LaVange, L

Woodcock, J. and LaVange, L. M. (2017). Master protocols to study multiple therapies, multiple diseases, or both. New England Journal of Medicine 377, 62--70

work page 2017
[51]

Ye, T., Shao, J., Yi, Y., and Zhao, Q. (2023). Toward better practice of covariate adjustment in analyzing randomized clinical trials. Journal of the American Statistical Association 118, 2370--2382

work page 2023
[52]

Yi, Y., Zhang, Y., Du, Y., and Ye, T. (2023). Testing for treatment effect twice using internal and external controls in clinical trials. Journal of Causal Inference 11, 20220018

work page 2023