Bayesian Design and Analysis of Precision Trials with Partial Borrowing
Shirin Golchi, Satoshi Morita · 2026
Read Bayesian Design and Analysis of Precision Trials with Partial Borrowing by Shirin Golchi, Satoshi Morita (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-10830.txt.
Introduction
Shirin Golchi and Satoshi Morita develop Bayesian design and analysis methods for precision trials that borrow partly from external data. Precision medicine needs methods for subgroup effects, where precise estimates are hard and sparse subgroups beg for outside information. Motivated by a gastric cancer trial, their model weights each external subject by fit with the target population, based on covariate patterns. Simulations compare the approach with dynamic borrowing. A design framework then takes priors from the external data to set decision boundaries and sample sizes.
Catalog record
| CREATOR | Shirin Golchi, Satoshi Morita |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 41970 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.10830
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Bayesian Design and Analysis of Precision Trials with Partial Borrowing Shirin Golchi, Satoshi Morita Abstract With the advancement of precision medicine there is an increasing need for design and analysis methods in clinical trials with the objective of investigating effect heterogeneity and estimating subgroup effects. As this requires precise estimation of interaction effects, borrowing information from external data sources including retrospective studies and early phase clinical trials to enrich the trial in sparse subgroups is pertinent. Motivated by a trial in gastric cancer we consider a practical design and analysis framework for borrowing from external data sources that only partially inform the inference. As the analysis model we propose an individually weighted model where the external data are weighted based on their fit with the target population based on the distribution of a set of covariates. In a simulation study we assess the performance of the model under various scenarios and make comparisons to dynamic borrowing. In addition, we provide a Bayesian design framework where design priors are extracted from the external data to determine decision boundaries and sample sizes. The design procedure is demonstrated within the context of our motivating example. Keywords: Bayesian operating characteristics; Individualized weights; External data; Power prior; Precision medicine; Subgroup effects. 1 Introduction Utilizing historical or external data to strengthen statistical inference in clinical trials has been explored in a variety of settings ( Dempster et al., 1982 ; Tarone, 1981 ; Ibrahim et al., 1998 ) . Information borrowing may be of interest for a variety of reasons including power gains in presence of sample size constraints in the context of rare diseases or pediatric trials ( Zocholl et al., 2026 ) , constructing or complementing the control group in single arm trials or trials with imbalanced randomization ratios in oncology and early phase drug development ( Thall & Simon, 1990 ; Zohar et al., 2008 ; Chen et al., 2016 ) . Depending on the objectives of the study, trial data may be augmented in one or more treatment groups. The statistical literature on external borrowing in clinical trials spans a spectrum of methods with the common objective of detecting and adjusting for the biases that may arise from discordance in information obtained from different sources. Particularly, several Bayesian methods have been proposed as the Bayesian framework offers a formal way of incorporating information via a prior distribution ( Neuenschwander et al., 2010 ; Schmidli et al., 2014 ; Rover & Friede, 2019 ; Lewis et al., 2019 ; Weber et al., 2019 ; Hobbs et al., 2011 ; Jiang et al., 2023 ) . Among the existing methods, power priors ( Chen & Ibrahim, 2000 ) , which discount external information by raising the likelihood constructed from the external data to a parameter, have enjoyed increased popularity due to their flexibility and generalizeability. The role of the power (or weight) parameter is to control the level of contribution of external information to inference and it may be specified in a static or dynamic manner ( Ibrahim et al., 2015 ; Duan, 2005 ; Duan et al., 2006 ; Carvalho & Ibrahim, 2021 ) . While The original formulation of the power prior relies on study level discounting, recent developments have extended the framework to accommodate partial borrowing ( Lu et al., 2022 ; Kwiatkowski et al., 2024 ; Alt et al., 2024 ) . Noteably, Alt et al., 2024 propose the latent exchangeability prior (LEAP) which assigns a power parameter to the individual level likelihoods. The individual level powers are then dynamically inferred from the data in a fully Bayesian model. With the advancement of precision medicine which aims to adapt treatment recommendations based on individual patient characteristics or biomarkers, inferring effect heterogeneity across subgroups and estimating subgroup specific effects is of primary or secondary interest in many clinical trials. Detecting effect heterogeneity requires estimating treatment-covariate interaction effects with sufficient precision. As small subgroup sizes are common in clinical trials, most trials are underpowered with respect to detection of effect heterogeneity and subgroup analysis. Therefore, borrowing information from external sources is particularly pertinent in the context of precision trials investigating effect heterogeneity and subgroup effects. While most of the work on external borrowing is focused on average treatment effects, recent developments have aimed to address subgroup analyses using external information. Daniells et al., 2025 consider a basket trial design where information is borrowed across baskets and from external sources via the exchangeablity-nonexchangeablity (EXNEX) model that incorporates a power prior in the non-exchangeability part of the mixture model. Schwartz et al., 2025 propose harmonized estimator for the subgroup specific effects that borrow from external control data but remains coherent with the average treatment effect estimate based on the current trial data. In this manuscript we consider information borrowing toward subgroup effects from external data sources that may inform differing subsets and functions of the model parameters. We propose an individually weighted formulation of the power prior where the power parameters are specified using a similarity measure inspired by the double-dipper similarity function proposed by Page & Quintana, 2018 based on the available covariates. We propose truncating the weights distribution to discard external data that score low according to this similarity measure where the truncation cut-off may be specified to limit the effective sample size of the external data. We assess the performance of the individually weighted model under a rage of simulation scenarios and make comparisons with LEAP as a representative of dynamic borrowing methods. In addition, we propose a Bayesian design procedure that builds upon the method of Psioda & Ibrahim, 2019 to specify design priors but accommodates the incomplete structure of the external data. We showcase the analysis and design framework with external borrowing within the context of a phase II randomized clinical trial in gastric cancer where retrospective and single arm external data are available to inform the treatment effect on the subgroup of patients with recurrent disease status. The remainder of this article is organized as follows. The gastric cancer trial example is presented as our motivating example in Section 2 . The individually weighted model for analysis with external borrowing is described in Section 3 within a precision trial framework from analysis and design perspectives. Section 4 follows with a simulation study to assess the model performance and make comparisons to dynamic borrowing. The proposed design and analysis methods are showcased in Section 5 in the context of the motivating example and a discussion follows in Section 6 . 2 Motivating Example The motivating clinical trial for the present work is a Randomized phase II study of S-1 plus cisplatin (SP) versus capecitabine plus cisplatin (XP) in patients with advanced or recurrent gastric cancer that cannot be curatively resected (XParTS-II) ( Nishikawa et al., 2018b ) . We consider the research question concerning the comparative effect of SP versus XP in the subgroup of patients with recurrent gastric cancer. However, this subgroup constitutes only 23% of patients enrolled in the trial and the small sample size does not allow for sufficient power for detecting treatment effect within this subgroup. To address this challenge we aim to use data from external sources to the trial to gain power toward