Adaptive randomization to improve utility-based dose-finding with bivariate ordinal outcomes.

Area of Science:

• Biostatistics
• Clinical Trial Design
• Pharmacometrics

Background:

• Optimizing experimental therapy dosing is crucial for balancing treatment efficacy and patient safety.
• Traditional clinical trial designs may not efficiently adapt to accumulating toxicity and efficacy data.
• Bivariate ordinal outcomes (toxicity, efficacy) present unique challenges in dose-finding studies.

Purpose of the Study:

• To propose a sequentially outcome-adaptive Bayesian design for dose selection of experimental therapies.
• To integrate patient utilities for bivariate ordinal outcomes (toxicity, efficacy) into an adaptive randomization framework.
• To control the risk of severe toxicity and identify optimal doses using posterior acceptability criteria.

Main Methods:

• A sequentially outcome-adaptive Bayesian design utilizing elicited utilities for toxicity and efficacy.
• Adaptive randomization based on posterior mean utilities and posterior probability of good outcomes.
• Saturated parametric models for marginal dose-toxicity/efficacy distributions and copula for joint distribution.
• Simulation-based prior mean computation and variance calibration for design optimization.

Main Results:

• The proposed design adaptively randomizes patients to doses with high posterior mean utilities.
• Near-optimality is defined by a nonincreasing utility increment relative to sample size.
• The method demonstrated effective dose selection in a simulated Phase I/II trial for pediatric brainstem gliomas.

Conclusions:

• The novel adaptive Bayesian design offers an efficient and safe approach for experimental therapy dose selection.
• This method effectively balances the competing risks of toxicity and the desire for efficacy.
• The design's utility in a pediatric brainstem glioma trial highlights its potential clinical applicability.