Master Protocols: Streamlining Clinical Trial Designs

Basket trials have emerged as a powerful tool in early-phase oncology drug development, enabling the evaluation of targeted therapies across multiple tumor types within a single study. Bayesian methods are widely used to facilitate adaptive information borrowing while maintaining statistical rigor. This talk reviews recent advancements in Bayesian basket trial designs and introduces a novel 3-component local power prior framework that offers modeling flexibility, computational efficiency, and explicit interpretability to facilitate cross-functional discussions. The framework incorporates global borrowing control, dynamic pairwise similarity assessment, and a threshold to restrict borrowing in highly heterogeneous settings. It also accommodates unequal sample sizes across baskets. Simulations demonstrate that the proposed approach achieves performance comparable to or better than established methods, including EXNEX and other MCMC-based approaches, while significantly reducing computational burden. We will also discuss practical considerations in tuning the borrowing parameters and illustrate how the proposed approach can be effectively implemented in early phase oncology trials. 

There has been an increasing trend in the development of targeted therapies. Platform trial designs have previously established efficiency in identifying efficacious treatments and delivering these options to patients quicker than traditional designs. Here we present an adaptive biomarker-driven platform trial design. In this setting, we have several candidate therapies each with a prospective predictive biomarker. As the trial progresses, we use a statistical model to estimate the efficacy of each drug/biomarker pair and use response adaptive randomization to preferentially assign patients to the optimal drug based on their entire biomarker profile. Compared to existing methods, this design has several advantages. First, it graduates efficacious treatments early and allows for ineffective treatments to be dropped, which produces better patient outcomes. Second, our methods can be used for biomarkers that are either binary or continuous. Finally, the flexibility of our design allows for Go/No Go decisions to advance treatments to either the total population or only to biomarker subgroups. We demonstrate the utility of our method through several simulation case studies. 

Strong evidence supports the effectiveness of pre-exposure prophylaxis (PrEP) in preventing human immunodeficiency virus infection; however, substantial provider- and patient-level barriers contribute to its underuse, disadvantaging racial, gender, and sexual minorities. Under a learning health system approach, a cluster-randomized trial evaluated the effect of linking a sexually transmitted infection testing bundle with a PrEP orderset to facilitate the prescription of PrEP. To rapidly assess this intervention, we seek novel methods to improve efficiency, and leveraging historical electronic health record data is a natural approach. However, current approaches to dynamic borrowing aren't able to account for the clustered nature of the data. We propose a method to dynamically borrow historical data from clinics where we expect similar trends in PrEP prescribing over time. We do this using a hierarchical Bayesian model with shrinkage via model averaging and the horseshoe prior. Our method results in increased efficiency relative to standard approaches, facilitating the rapid evaluation of this intervention and working to improve PrEP access in historically disadvantaged populations. 

A master protocol is a comprehensive document that outlines the plan for clinical research within multiple sub-studies. As with any protocol, it includes detailed information on the objectives, design, methodology, statistical considerations, and organization. Regardless of the protocol type, clinical trials must begin with the fundamental design question "What is the question we want to answer?" followed by the question "What do we need to inform decision making?". For master protocols these questions are considered not only for a single asset or patient population but also for an indication or a therapeutic area. When developing a master protocol, study teams tend to focus on their single asset rather than planning holistically across an indication or therapeutic area resulting in a stitched together "Frankenstein's monster" protocol. By way of examples, this paper will speak to questions, decisions, and design elements to establish a framework that informs the trial design described by the master protocol and its appendices/addendums. 

Combination drug therapies hold significant promise in enhancing treatment efficacy, particularly in fields such as oncology, immunotherapy, and infectious diseases. Designing clinical trials for these regimens poses unique challenges due to multiple hypothesis testing, shared control groups, and overlapping treatment components that induce complex correlation structures. In this paper, we develop a novel statistical framework tailored for early-phase translational combination therapy trials, with a focus on platform trial designs. Our methodology introduces a generalized Dunnett's procedure that controls false positive rates by accounting for the correlations between treatment arms. Additionally, we propose strategies for power analysis and sample size optimization that leverage preclinical data to estimate effect sizes, synergy parameters, and inter-arm correlations. Simulation studies demonstrate that our approach not only controls various false positive metrics under diverse trial scenarios but also informs optimal allocation ratios to maximize power. A real-data application further illustrates the integration of translational preclinical insights into the clinical trial design process. An open-source R package is provided to support the application of our methods in practice. Overall, our framework offers statistically robust guidance for the design of early-phase combination therapy trials, aiming to enhance the efficiency of the bench-to-bedside transition.