Pioneering Bayesian Dynamic Borrowing: Enhancing Clinical Trial Design and Analysis

Zizhong Tian Chair
Penn State University
 
Monday, Aug 4: 2:00 PM - 3:50 PM
4072 
Contributed Papers 
Music City Center 
Room: CC-208A 

Main Sponsor

Biopharmaceutical Section

Presentations

A Bayesian hybrid Dynamic Borrowing Framework Incorporating Covariates

In early-phase drug development, the goal is to assess whether a novel agent adds activity to a monotherapy. A Bayesian hybrid design with dynamic borrowing from historical monotherapy data offers a robust approach to enhance study efficiency and decision-making. Lu et al. (2024) proposed a dynamic borrowing framework for binary outcomes based on the dynamic power prior (DPP) and the similarity of outcomes between the study control and historical control. This approach is more robust than traditional single-arm designs and can significantly improve statistical power at the design stage. At the analysis stage, adjusting for important covariates improves modeling by controlling for variability and confounding effects. In this work, we develop an analysis framework that incorporates covariates through propensity scores into the DPP (PS-DPP), enabling more informed borrowing by accounting for the similarity between historical and current control data based on both covariates and outcomes. Simulation studies show that PS-DPP improves analytical performance, especially when there are substantial differences in both response rates and covariates between historical and current controls. 

Keywords

Bayesian hybrid design

dynamic borrowing

nonconcurrent control

power prior

propensity score 

Co-Author(s)

Yiyuan Huang
Philip He, Daiichi Sankyo Inc.

First Author

Zhaohua Lu

Presenting Author

Zhaohua Lu

Boosting the Power of Hybrid Design When Crossover Occurred in Historical Control

The hybrid design of augmenting concurrent control by borrowing historical control offers a promising solution when enrolling control patients is challengeable. It not only ensures the desired trial efficiency but also maintains the randomization property of clinical trials. In this design, population heterogeneity between historical and concurrent controls limits the data that can be borrowed, thereby hindering trial efficiency. Many studies addressed this issue focusing on two sources of heterogeneity: differences in patients' characteristics, and population drift caused by evolving SOC. However, when overall survival is the endpoint in oncology trials, crossover occurred in historical control result in additional discrepancies between historical and concurrent controls. Existing methods are deficient to fully resolve this issue.
Our study proposed a new design by integrating the adjustment of crossover effect into a hybrid framework of Bayesian dynamic borrowing model. We conducted extensive simulations to evaluate the performances under various scenarios, demonstrating the effectiveness of this framework in reducing estimate bias and boosting the power of hybrid design. 

Keywords

Hybrid design

RCT

Historical control

Crossover

Bayesian dynamic borrowing

Commensurate prior 

Co-Author(s)

Qixiang Xu, Department of Biostatistics, Yale University
Qing Liu, Amgen Inc.

First Author

Leiwen Gao, Amgen Inc.

Presenting Author

Leiwen Gao, Amgen Inc.

WITHDRAWN Building a Hybrid External Control Arm Using a Fusion of Bayesian Borrowing and Causal Inference

Hybrid external control arms (HECAs) are an innovative study design that augment a randomized control arm with data from a historical control arm. HECAs have the potential to reduce patient burden, trial costs, and improve early decision-making while maintaining the scientific rigor of randomized control arms. This presentation will explore the construction and analysis of HECAs. One possible approach for analyzing HECAs will be explored in depth, which leverages Bayesian borrowing and causal inference to address challenges in patient exchangeability and outcome heterogeneity. In particular, techniques such as inverse probability of treatment weighting (IPTW) and robust mixture priors are used to ensure doubly robust estimation. A case study will demonstrate the application of these methods using data from a phase II trial, validating the HECA approach against frequentist benchmarks. This innovative methodology holds significant promise for advancing precision and efficiency in clinical trial design and analysis. 

Keywords

hybrid external control arms (HECAs)

Bayesian borrowing

Causal inference

inverse probability of treatment weighting (IPTWs)

robust mixture priors

clinical trial design 

Co-Author

Antara Majumdar, GSK

First Author

Ilana Trumble, GSK

Concordance-based prior to dynamically borrow information for pediatric extrapolation

In pediatric drug development, effectively borrowing information from adult trials can significantly reduce sample size and improve trial efficiency while maintaining robust inference. We propose a novel Bayesian dynamic borrowing approach that adjusts the amount of information borrowed from adult populations based on the concordance of a clinical endpoint and a predictive biomarker between the two populations. Our method leverages both clinical and biomarker data to guide borrowing decisions, balancing the concordance and divergence observed in different endpoints. Through simulation studies and real clinical data examples, we demonstrate that our approach consistently improves estimation accuracy and power while maintaining appropriate type I error. The proposed framework has broad applications in regulatory settings where adaptive borrowing strategies are crucial for ethical and efficiency reasons. 

Keywords

Bayesian data borrowing

Pediatric extrapolation

Clinical trial 

Co-Author(s)

Siyu Zhang, Vertex Pharmaceuticals
Pengyu Liu
Fengjuan Xuan, Vertex

First Author

Weiyu Zhou

Presenting Author

Weiyu Zhou

Matching-assisted power prior for incorporating real-world data in randomized clinical trial

Leveraging external data to supplement randomized clinical trials has become increasingly popular, particularly for medical device and drug discovery. In rare diseases, recruiting enough patients for large-scale trials is challenging. To address this, small hybrid trials can borrow historical controls or real-world data (RWD) to increase statistical power, but borrowing must follow a statistically principled manner. This paper proposes a matching-assisted power prior method to mitigate bias when incorporating external data. Using template matching, a subset of comparable external subjects is grouped and assigned weights based on their similarity to the current study population. These weighted groups are then integrated into Bayesian inference through power priors. Unlike traditional power prior methods, which apply similar discounts to all control patients, our approach pre-selects high-quality controls, improving the reliability of borrowed data. Through simulation studies, we compare its performance with the propensity score-integrated power prior approach. Finally, we demonstrate its practical implementation using data from a real acupuncture clinical trial. 

Keywords

Bayesian dynamic borrowing

power prior

template matching

propensity score

real-world data 

Co-Author(s)

Biqing Yang, The Ohio State University
Xinyi Xu
Bo Lu, The Ohio State University

First Author

Ruoyuan Qian, The Ohio State University

Presenting Author

Ruoyuan Qian, The Ohio State University

Propensity score stratified MAP prior and posterior inference for incorporating external data

Incorporation of external information is becoming increasingly common when designing clinical trials. Availability of multiple sources of information has inspired the development of methodologies that account for potential heterogeneity not only between the prospective trial and the pooled external data sources but also between the different external data sources themselves. Our approach proposes an intuitive way of handling such a scenario for the continuous outcomes setting by using propensity score-based stratification and then utilizing robust meta-analytic predictive priors for each stratum to incorporate the prior data to distinguish among different external data sources in each stratum. Through extensive simulations, our approach proves to be more efficient and less biased than the currently available methods. A real case study using clinical trials that study schizophrenia from multiple different sources is also included. 

Keywords

propensity score

Bayesian borrowing

external controls

heterogeneity 

Co-Author(s)

Dooti Roy, Boehringer Ingelheim Pharmaceuticals, Inc.
Zheng Zhu
Martin Oliver Sailer, Boehringer Ingelheim

First Author

Angela Zhu, Boehringer Ingelheim

Presenting Author

Angela Zhu, Boehringer Ingelheim