Bayesian phase I/II adaptively randomized oncology trials with combined drugs

Ying Yuan, Guosheng YinThe Annals of Applied Statistics 2011 · DOI

TL;DR

This paper introduces a novel integrated Phase I/II clinical trial design for drug combinations in oncology. It combines a Bayesian approach for identifying safe and effective dose combinations with an adaptive randomization strategy that efficiently allocates patients to the most promising treatments, aiming to accelerate drug development and improve patient outcomes.

What Problem Does This Paper Address?

Developing new cancer treatments often involves combining multiple drugs to achieve synergistic effects or overcome resistance. However, testing all possible dose combinations is complex and resource-intensive. Traditional clinical trials separate Phase I (safety and dose-finding) and Phase II (efficacy testing), leading to delays. Furthermore, drug combination trials face unique challenges: the dose-toxicity relationship is not always monotonic, and multiple combinations might exhibit similar toxicity profiles, making it difficult to select the best candidates for Phase II. This paper addresses the need for a more efficient, integrated approach that can simultaneously explore safety and efficacy in drug combination trials, specifically by improving how patients are assigned to treatment arms in Phase II based on emerging efficacy data.

Methods

The proposed design integrates two key components:

  1. Bayesian Copula-Type Model for Dose Finding (Phase I): The authors employ a Bayesian copula-type regression model to characterize the joint toxicity probability of drug combinations. This model allows for flexibility in capturing complex dose-toxicity relationships and leverages prior information about individual drug toxicities. It reduces to the well-known Continual Reassessment Method (CRM) for single agents when only one drug is considered. This phase aims to identify a set of "admissible" dose combinations that meet predefined safety criteria.
  2. Moving-Reference Adaptive Randomization (MAR) (Phase II): Once admissible doses are identified, Phase II begins. Patients are adaptively randomized to these arms. The novel MAR scheme is designed to assign patients to more efficacious arms with higher probability. Unlike fixed-reference methods, MAR uses a "moving reference" that adapts as the trial progresses. It compares each arm not to a single fixed reference, but to the current average efficacy of the remaining arms. This "zoomed-in" comparison allows for higher resolution in distinguishing between arms with similar efficacy, leading to more efficient allocation of patients to superior treatments. The paper also extends this to group sequential adaptive randomization for cases where efficacy is late-onset.

Datasets

The paper uses a simulated melanoma clinical trial as a motivating example. No specific public datasets were used for training or evaluation; instead, the authors generated data through extensive simulation studies to assess the operating characteristics of their proposed design.

Key Insights and Ideas

  • Seamless Integration: The design elegantly bridges Phase I and Phase II, allowing for immediate transition and utilization of data from both phases.
  • Adaptive Randomization for Drug Combinations: The MAR scheme is a significant advancement for drug combination trials, addressing the challenge of comparing multiple arms with potentially similar efficacy profiles.
  • Moving Reference Advantage: The adaptive nature of the MAR reference is crucial. It allows for more nuanced comparisons, especially when efficacy differences are subtle or when some arms are clearly inferior. This contrasts with fixed-reference methods that can struggle to differentiate between arms once they are all performing better than the fixed reference.
  • Efficiency in Patient Allocation: By accounting for both the point estimate and variability of efficacy, MAR aims to allocate more patients to truly superior treatments, maximizing the chance of identifying the best drug combination and potentially leading to better patient outcomes.
  • Robustness: The design is shown to be robust to prior specification and model misspecifications through sensitivity analyses.

Key Results

  • Melanoma Trial Illustration: The proposed design was applied to a simulated melanoma trial, demonstrating its practical application and the dynamic nature of adaptive randomization probabilities. The trial successfully identified a desirable dose combination.
  • Simulation Studies: Extensive simulations showed that the MAR procedure allocated a substantially higher proportion of patients to the most efficacious arm compared to fixed-reference adaptive randomization (FAR) across various scenarios. For instance, in one scenario, MAR assigned 58.5% of patients to the best arm, while FAR assigned only 40.6%.
  • Admissible Dose Selection: In Phase I, the design generally selected a high percentage of true admissible doses (often >90%) and provided a reasonable estimate of the number of admissible doses for Phase II.
  • Late-Onset Efficacy: The group sequential AR approach for late-onset efficacy showed that while larger group sizes could shorten trial duration, they might slightly reduce the number of patients allocated to the target dose.

Limitations and Future Work

  • Binary Endpoints: The current design primarily assumes binary toxicity and efficacy outcomes. The authors suggest future work to extend the model to handle ordinal or continuous endpoints (e.g., multiple toxicity grades, time-to-event data) by incorporating techniques like quasi-binomial likelihoods or survival models.
  • Assumed Observability: The core design assumes that efficacy outcomes are observable quickly enough for real-time decision-making. The group sequential AR addresses this for late-onset efficacy, but the optimal choice of group size and its impact on trial duration versus patient allocation remains a consideration.
  • Computational Implementation: The paper mentions the availability of an implemented C++ executable, but further validation and broader adoption would be beneficial.
  • Trade-off Management: While the paper focuses on efficacy-driven randomization, future work could explore more explicit methods for balancing toxicity and efficacy trade-offs, potentially using odds ratios as a measure of desirability.

Summary updated

Abstract

We propose a new integrated phase I/II trial design to identify the most efficacious dose combination that also satisfies certain safety requirements for drug-combination trials. We first take a Bayesian copula-type model for dose finding in phase I. After identifying a set of admissible doses, we immediately move the entire set forward to phase II. We propose a novel adaptive randomization scheme to favor assigning patients to more efficacious dose-combination arms. Our adaptive randomization scheme takes into account both the point estimate and variability of efficacy. By using a moving reference to compare the relative efficacy among treatment arms, our method achieves a high resolution to distinguish different arms. We also consider groupwise adaptive randomization when efficacy is late-onset. We conduct extensive simulation studies to examine the operating characteristics of the proposed design, and illustrate our method using a phase I/II melanoma clinical trial.
BibTeX
@article{yuan2011bayesian,
  author = {Ying Yuan and Guosheng Yin},
  title = {{Bayesian phase I/II adaptively randomized oncology trials with combined drugs}},
  journal = {The Annals of Applied Statistics},
  year = {2011},
  doi = {10.1214/10-aoas433},
}
Figure