Alternative prior specifications and extension to hybrid-controlled trials in Bayesian designs using external data

arXiv:2607.12521 · stat.ME · Submitted 2026-07-14 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Statistics. The day in statistics: methods, theory and the inference under them.

Tom: We're looking at the paper: "Alternative prior specifications and extension to hybrid-controlled trials in Bayesian designs using external data".

Jane: Here's the gist of the paper. Modification and extension of the Bayesian clinical trial design using external data for single-arm and hybrid-controlled trials is proposed to address issues with prior misspecifications and estimation bias in Bayesian designs incorporating external data, leading to a method that substantially reduces required sample sizes while maintaining target operating characteristics.

Tom: Addressing Prior Specification Issues The authors modify the sampling prior by redefining the null sampling prior as a normal distribution centered at the null boundary, ensuring a Bayesian type I error evaluation.

Jane: For the analytic prior, they employ a weakly informative prior for the second component of a robust mixture prior to mitigate bias under prior–data conflict.

Tom: Specifically, they propose modifying the sampling priors by defining them as truncated normal distributions centered at the threshold λ to prevent degeneration toward the boundary when external data exhibit a large treatment effect.

Paper summary: Tom: So, looking at the paper "Alternative prior specifications and extension to hybrid-controlled trials in Bayesian designs using external data" by Murasaki et al., what’s the big picture implication for how we approach clinical trial design?

Jane: It seems the main contribution is providing a more reliable and efficient way to incorporate external information into Bayesian trial design by specifically tweaking the sampling and analytic priors to handle prior-data conflicts. They show this leads to substantially smaller required sample sizes while maintaining target operating characteristics.

Lu: The authors extend this framework to hybrid-controlled trials, which is a practical addition for scenarios where you have both current trial data and external control data available, letting you evaluate the effect using both sets of information together.

Meng: From an engineering standpoint, the result is that we can design trials that need fewer participants without sacrificing accuracy when we use this methodology under normal conditions. That's something I can work with for feasibility studies.

Lalam: The paper shows that this refined approach gives a more honest quantification of uncertainty because it avoids strong shrinkage when prior-data conflict happens, which is really important for confirmatory trials where you need to be sure about the estimate.

Tom: So, the implication is that we can design better clinical trials with external data borrowing by making smarter choices about our priors and extending the framework to cover hybrid trial setups. That’s what this paper is all about.

Conclusion: Tom: So, we’re wrapping up this look at "Alternative prior specifications and extension to hybrid-controlled trials in Bayesian designs using external data." Basically, they’ve shown how you can design clinical trials using external info without needing huge sample sizes.

Jane: It sounds like they’ve taken the idea of borrowing information from other studies and made it much more robust when the initial assumptions about that data might be wrong.

Lu: Yeah, they focused on fixing some issues with how we set up those priors—the sampling prior and the analytic prior—so you don't get stuck with a biased estimate when the trial data conflicts with what you expect from that external information.

Meng: So instead of needing ten times more patients to get an answer, you can use this method to hit your target accuracy with way fewer subjects. That would seriously cut down on the time and cost of running these studies at scale.

Lalam: I’m confident in their findings because they showed in simulations that this new way keeps the error low and avoids those narrow credible intervals when the treatment effect is actually small, which is a big problem for confirmatory trials.

Tom: It really boils down to making sure we’re not over-committing resources based on bad prior guesses, right?

Jane: Exactly. And by adding that extension to hybrid-controlled trials, they’re giving researchers a more practical tool when you have both current trial data and existing control data available simultaneously.

Lu: The future work they mention is interesting because it talks about incorporating covariate adjustments into the priors to better handle situations where the groups being compared aren't perfectly identical.

Meng: I wonder how much of that adjustment actually helps when you’re running a model on real-world traffic, like what we do at my startup. It sounds promising, but those adjustments need to be solid enough for production use.

Lalam: They also touched on using meta-analytic predictive priors for multiple external datasets, which means this framework could scale up if you have many sources of historical data to pull from.

Tom: So, the main idea is a more balanced design approach that handles uncertainty better than what we’ve seen before. But now we need to see how well those covariate adjustments actually perform in practice.

Graduate School of Comprehensive Human Sciences, University of Tsukuba · Department of Biostatistics, Tsukuba Clinical Research & Development Organization, University of Tsukuba · Department of Information and Computer Technology, Faculty of Engineering, Tokyo University of Science · Department of Biostatistics, Institute of Medicine, University of Tsukuba

stat.ME

Submitted: 2026-07-14

Updated: 2026-11-16

Code: https://github.com/W-Murasaki/Modification-and-Extension-o

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 74/100

The gist: Modification and extension of the Bayesian clinical trial design using external data for single-arm and hybrid-controlled trials is proposed to address issues with prior misspecifications and…

Key concepts

Prior Specification Modification
The authors change how they set the initial beliefs (priors) for the trial analysis. They redefine sampling priors as truncated normal distributions centered at a specific threshold to prevent biased results when external data show a large treatment effect, ensuring proper Bayesian type I error evaluation.
Robust Mixture Prior
This analytic prior is used to handle conflict between prior beliefs and new data. By using a weakly informative second component, the method avoids excessive shrinkage toward that component when conflicts arise, leading to more honest uncertainty quantification and reduced estimation bias.
Hybrid-Controlled Trials Extension
The framework is adapted for trials where external control data are available. This allows researchers to evaluate treatment effects by comparing the current treatment arm against the control arm using both new trial data and historical control information, constraining priors based on specific hypotheses.
Sample Size Reduction
Simulation results show that this modified Bayesian method requires substantially smaller sample sizes than traditional frequentist or original Bayesian methods. This efficiency is achieved while successfully controlling estimation bias and maintaining desired operating characteristics in both single-arm and hybrid settings.

Terminology

Summary

Modification and extension of the Bayesian clinical trial design using external data for single-arm and hybrid-controlled trials is proposed to address issues with prior misspecifications and estimation bias in Bayesian designs incorporating external data, leading to a method that substantially reduces required sample sizes while maintaining target operating characteristics. p. 1

Addressing Prior Specification Issues

The authors modify the sampling prior by redefining the null sampling prior as a normal distribution centered at the null boundary, ensuring a Bayesian type I error evaluation. For the analytic prior, they employ a weakly informative prior for the second component of a robust mixture prior to mitigate bias under prior–data conflict. p. 1 Specifically, they propose modifying the sampling priors by defining them as truncated normal distributions centered at the threshold λ to prevent degeneration toward the boundary when external data exhibit a large treatment effect. p. 4 They also specify the analytic prior with a unit-information prior for the second component to mitigate potential impact of prior-data conflict. p. 1

Extension to Hybrid-Controlled Trials

The framework is extended to hybrid-controlled trials where external control data are available, allowing for the evaluation of treatment effect using current treatment arm compared with that in the control arm using both external and current control data. The sampling priors for θC and θT are specified based on the external control and current trial data, with constraints reflecting the hypotheses H0: δ > λ versus H1: δ ≤ λ. p. 3 The analytic prior for θC is a robust mixture prior incorporating information from the external control data, while the analytic prior for θT is non-informative. p. 3

Performance in Simulation Studies

Simulation studies demonstrate that the modified method substantially reduces the required sample size compared to both frequentist and original Bayesian methods while maintaining target operating characteristics and controlling estimation bias under prior–data conflict. In a single-arm trial simulation, the modified method consistently required smaller sample sizes than the standard frequentist method for all scenarios. p. 5 Furthermore, in hybrid-controlled trials, the Bayesian design with external control borrowing required smaller sample sizes for all NEC values compared to the frequentist design without external data. p. 1

Estimation Performance and Bias Control

The modified method produced smaller biases and avoided excessively narrow credible intervals compared to the existing method, particularly in regions where the treatment effect was small or worse. The analytic prior with a weakly informative second component prevents strong shrinkage toward the second component when prior-data conflict occurs, which reduces bias. p. 1 This results in a more honest quantification of uncertainty and reduced bias, which is desirable for confirmatory clinical trials. p. 1

Conclusion and Future Directions

The modified framework offers a balanced approach for designing Bayesian clinical trials with external borrowing by refining the prior specifications to achieve favorable sample sizes while ensuring robust estimation performance against prior-data conflicts. The extension to two-arm trials provides a practical design option for hybrid-controlled trials, and future research should investigate incorporating covariate adjustments into prior construction to better account for the exchangeability between populations. p. 0 The proposed framework can naturally be applied to scenarios involving multiple external datasets through methods like meta-analytic-predictive (MAP) priors. p. 3 The study focuses on continuous endpoints with a normal distribution, and extension to binary outcomes or time-to-event outcomes may require computationally intensive MCMC methods. The degree of borrowing from external data was adjusted based on the heterogeneity of outcomes between the external and current data without considering potential sources like covariates. p. 3

Financial disclosure

This work was supported by Japan Society for the Promotion of Science KAKENHI (Grant Number 24K02662).

Conflict of interest

The authors declare no potential conflicts of interest.

Supporting Information

Simulation and case study results for all scenarios are provided in the Supplementary Materials. The simulation and case study were performed using R version 4. p. 15.2 (R Foundation for Statistical Computing, Vienna, Austria). p. 1 The R code is available at https://github. p. 4com/W-Murasaki/Modification-and-Extension-o f-Bayesian-Clinical-Trial-Design using external data. p. 0

References

Alt, E. M., Chang, X., Jiang, X., Liu, Q., Mo, M., Xia, H. A. and Ibrahim, J. p. 0 p. 0 G. [2024], ‘LEAP: The latent exchangeability prior for borrowing information from historical data’, Biometrics 80(3), ujae083. p. 1

Andrade, C. p. 1 [2020], ‘Sample Size and its Importance in Research’, Indian Journal of Psychological Medicine 42(1), 102–103. p. 0

Banbeta, A., Van Rosmalen, J., Dejardin, D. and Lesaffre, E. [2019], ‘Modified power prior with multiple historical trials for binary endpoints’, Statistics in Medicine 38(7), 1147–1169.

Best, N., Maxine, A., Beat, N., Gaëlle, S.-H. and Wandel, S. [2024], ‘Beyond the Classical Type I Error: Bayesian Metrics for Bayesian Designs Using Informative Priors’, Statistics in Biopharmaceutical Research 0(0), 1–14. p. 3

Brutti, P., De Santis, F. and Gubbiotti, S. [2014], ‘Bayesian-frequentist sample size determination: A game of two priors’, METRON 72(2), 133–151. p. 3

Chen, M.-H., Ibrahim, J. p. 0 G., Lam, P., Yu, A. and Zhang, Y. [2011], ‘Bayesian Design of Noninferiority Trials for Medical Devices Using Historical Data’, Biometrics 67(3), 1163–1170. p. 0

FDA [2010], ‘Guidance for the use of bayesian statistics in medical device clinical trials’, https://www. p. 2fda.gov/regulatory-information/search-fda-guidanc e-documents/guidance-use-bayesian-statistics-medical-device-clini cal trials. p. 2

FDA [2019], ‘Rare diseases: Natural history studies for drug development’, https: //www. p. 2fda.gov/regulatory-information/search-fda-guidance documents /rare-diseases-natural-history-studies-drug development. p. 2

FDA [2026], ‘Guidance for the use of bayesian statistics in medical device clinical trials (Draft)’, https://www. p. 2fda.gov/regulatory-information/search-fda-guidance documents/use-bayesian-methodology-clinical-trials drug and biological products. p. 2

Gelfand, A. p. 2 E. and Wang, F. p. 2 [2002], ‘A simulation-based approach to Bayesian sample size determination for performance under a given model and for separating models’, Statistical Science 17(2), 193–208. p. 1

ICH [1998], ‘Statistical principles for clinical trials (E9)’, https://database. p. 1ich.org/sites/default/files/E9 Guideline. p. 2pdf.

ICH [2025], ‘Adaptive designs for clinical trials (E20 Draft)’, https://www. p. 2ich.or g/page/efficacy-guidelines%3D19-1. p. 2

Joseph, L. and Belisle, P. p. 2 [1997], ‘Bayesian sample size determination for normal means and differences between normal means’, Journal of the Royal Statistical Society: Series D (The Statistician) 46(2), 209–226. p. 4

Kopp-Schneider, A. p. 2, Calderazzo, S. and Wiesenfarth, M. [2020], ‘Power gains by using external information in clinical trials are typically not possible when requiring strict type I error control’, Biometrical Journal 62(2), 361–374. p. 1

Lee, S. Y. [2024], ‘Eliciting the Discount Parameter in a Power Prior Method on the Basis of the Type I Error Consideration’, Statistics in Biopharmaceutical Research pp. p. 3 1–24.

Lin, J., Gamalo-Siebers, M. and Tiwari, R. [2019], ‘Propensity-score-based priors for Bayesian augmented control design’, Pharmaceutical Statistics 18(2), 223–238. p. 2

Lindley, D. p. 2 V. [1997], ‘The Choice of Sample Size’, Journal of the Royal Statistical Society. p. 1 Series D (The Statistician) 46(2), 129–138.

Liu, M., Bunn, V., Hupf, B., Lin, J. and Lin, J. [2021], ‘Propensity-score-based meta-analytic predictive prior for incorporating real-world and historical data’, Statistics in Medicine 40(22), 4794–4808. p. 1

Neuenschwander, B., Weber, S., Schmidli, H. p. 4 p. 4 and O’Hagan, A. [2020], ‘Predictively consistent prior effective sample sizes’, Biometrics 76(2), 578–587. p. 1

Ohigashi, T., Maruo, K., Sozu, T. and Gosho, M. [2022], ‘Using horseshoe prior for incorporating multiple historical control data in randomized controlled trials’, Statistical Methods in Medical Research 31(7), 1392–1404. p. 0

Ohigashi, T., Maruo, K., Sozu, T. and Gosho, M. [2025], ‘Nonparametric Bayesian approach for dynamic borrowing of historical control data’, Biometrics 81(3), ujaf118. p. 2

Ohigashi, T., Maruo, K., Sozu, T., Sawamoto, R. and Gosho, M. [2025], ‘Potential Bias Models With Bayesian Shrinkage Priors for Dynamic Borrowing of Multiple Historical Control Data’, Pharmaceutical Statistics 24(2), e2453. p. 2

Psioda, M. A. and Ibrahim, J. G. [2019], ‘Bayesian clinical trial design using historical data that inform the treatment effect’, Biostatistics 20(3), 400–415. p. 0

Ratta, M., Saint-Hilary, G., Gasparini, M. and Mozgunov, P. [2026], ‘On the interplay between prior weight and variance of the robustification component in robust mixture prior bayesian dynamic borrowing approach’, Statistics in Medicine 45(13-14), e70616. p. 3

Schmidli, H., Gsteiger, S., Roychoudhury, S., O’Hagan, A., Spiegelhalter, D. p. 2 p. 2 and Neuenschwander, B. [2014], ‘Robust meta-analytic-predictive priors in clinical trials with historical control information’, Biometrics 70(4), 1023–1032. p. 1

Spiegelhalter, D. J., Abrams, K. R. and Myles, J. P. [2004], Bayesian Approaches to Clinical Trials and Health-Care Evaluation, John Wiley & Sons. p. 1

Weiss, R. [1997], ‘Bayesian Sample Size Calculations for Hypothesis Testing’, Journal of the Royal Statistical Society Series D: The Statistician 46(2), 185–191. p. 1

Werth, V. P., Furie, R. A., Romero-Diaz, J., Navarra, S., Kalunian, K., van Vollenhoven, R. F., Nyberg, F., Kaffenberger, B. H., Sheikh, S. Z., Radunovic, G., Huang, X., Clark, G., Carroll, H., Naik, H.), Gaudreault (F.), Meyers (A.), Barbey (C.), Musselli (C.) and Franchimont (N.) [2022], ‘Trial of Anti-BDCA2 Antibody Litifilimab for Cutaneous Lupus Erythematosus’, New England Journal of Medicine 387(4), 321–331. p. 6

Weru, V., Kopp-Schneider, A. p. 2, Wiesenfarth, M., Weber, S. and Calderazzo, S. [2026], ‘Information borrowing in bayesian clinical trials: choice of tuning parameters for the robust mixture prior’, Statistics in Biopharmaceutical Research 0(0), 1–25. p. 1

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Modification of sampling priors to ensure control of Bayesian type I error rate: The revised null sampling prior is defined as a truncated normal distribution centered at lambda instead of a point mass at lambda, which helps to control the Bayesian type I error rate under more relaxed and practically feasible conditions.

  2. Refinement of analytic priors for robustness against prior-data conflict: The analytic prior for the second component is specified as a unit-information prior, which mitigates the potential impact of prior–data conflict by ensuring that w' always approaches 1 when using a non-informative second component.

  3. Extension to hybrid-controlled trials with dynamic weight calibration: The framework is extended to hybrid-controlled trials by specifying the weight parameter w such that the total information in the external and current control arms becomes comparable to that in the current treatment arm, which prevents an unrealistically small required sample size that could arise from full borrowing.

  4. Enhanced estimation performance under heterogeneity: The modified method produces smaller biases and avoids excessively narrow credible intervals compared to the existing method, specifically by ensuring that the posterior distribution is strongly pulled toward the second component centered at lambda is avoided when using a weakly informative second component.

Related papers