Potential outcome simulation for efficient head-to-head comparison of adaptive dose-finding designs

Area of Science:

• Pharmacometrics
• Biostatistics
• Clinical Trial Design

Background:

• Dose-finding trials are crucial for drug development, requiring statistical designs to guide dosing decisions.
• Evaluating competing dose-finding designs typically involves time-consuming large-scale simulation studies.
• Current simulation methods limit the scope due to extensive computational requirements.

Purpose of the Study:

• To introduce a more efficient simulation approach for designing and evaluating dose-finding trials.
• To enable faster and more comprehensive head-to-head comparisons of different statistical designs.
• To reduce the computational burden and Monte Carlo error in simulation studies.

Main Methods:

• Simulating all potential individual outcomes at each dose level in advance.
• Applying pre-simulated datasets to multiple competing dose-finding designs for comparison.
• Utilizing advance-simulated datasets for assessing design performance across various configurations.

Main Results:

• Demonstrated substantial reductions in Monte Carlo error for comparing design performance metrics.
• Showcased efficiency gains, with one case study requiring a 48-times smaller simulation study.
• Highlighted the reusability of advance-simulated datasets for multiple configurations.

Conclusions:

• The new simulation approach offers significant efficiency gains for dose-finding trial evaluations.
• Researchers are encouraged to adopt this method for more robust and efficient simulation studies.
• The R package 'escalation' has been updated to support the implementation of this new approach.