Urn models for response-adaptive randomized designs: a simulation study based on a non-adaptive randomized trial.

Area of Science:

• Biostatistics
• Clinical Trial Design
• Medical Statistics

Background:

• Response-adaptive designs enhance randomized clinical trials ethically and economically.
• Urn models, using virtual urns with colored balls, guide treatment assignment probabilities dynamically.
• These models adapt assignments based on accumulating trial data.

Purpose of the Study:

• To propose and evaluate Randomly Reinforced Urn (RRU) models for response-adaptive randomization.
• To compare RRU design performance against non-adaptive approaches in a clinical trial simulation.
• To provide R software code for practical implementation of the RRU design.

Main Methods:

• Simulation of 10,000 trials using the RRU model across varied sample sizes.
• Application of the RRU model to a published trial on home enteral nutrition in cancer patients.
• Sensitivity analysis on different urn compositions and comparison with non-adaptive designs.

Main Results:

• The RRU design allocated fewer patients to the inferior treatment in approximately 75% of simulations.
• Empirical power of the t-test for treatment effect remained comparable between RRU and non-adaptive designs.
• Simulations were based on a real-world randomized clinical trial.

Conclusions:

• Randomly Reinforced Urn models offer a more ethical and potentially cost-effective alternative to non-adaptive designs.
• The RRU design effectively minimizes patient exposure to less effective treatments.
• Statistical power is maintained, making RRU a viable option for adaptive clinical trials.