A benchmark for dose-finding studies with unknown ordering.

Area of Science:

• Clinical Trial Design
• Biostatistics
• Pharmacology

Background:

• The nonparametric optimal benchmark is crucial for evaluating dose-finding designs, offering an upper performance bound.
• A key assumption is the monotonic toxicity ordering of doses, which is often challenging in combination studies.
• Existing benchmarks do not fully address the uncertainty inherent in ordering dose combinations.

Purpose of the Study:

• To propose a generalized nonparametric optimal benchmark that incorporates uncertainty in dose ordering.
• To provide a sharper upper bound on the performance of dose-finding designs in complex scenarios.
• To enhance the evaluation of dose-finding designs for Phase I and Phase I/II combination trials.

Main Methods:

• Developed a generalized benchmark that quantifies the probability of each dose ordering based on patient data.
• The approach accommodates trials with multiple endpoints, including discrete and continuous outcomes.
• Applied the benchmark to evaluate novel dose-finding designs for combination therapies.

Main Results:

• The proposed generalized benchmark offers a more accurate and sharper upper bound on design performance.
• It effectively accounts for the uncertainty in dose-toxicity ordering, a common issue in combination trials.
• Demonstrated utility in Phase I (binary toxicity) and Phase I/II (binary toxicity, continuous efficacy) combination trials.

Conclusions:

• The generalized benchmark provides a more robust tool for assessing dose-finding designs in complex clinical trials.
• This method improves the evaluation of combination therapies by addressing ordering uncertainties.
• The approach is versatile, applicable to various endpoints and trial phases.