A NOVEL FRAMEWORK TO ESTIMATE MULTIDIMENSIONAL MINIMUM EFFECTIVE DOSES USING ASYMMETRIC POSTERIOR GAIN AND -TAPERING.

Area of Science:

• Biostatistics
• Clinical Trial Design
• Behavioral Interventions

Background:

• Dose-finding clinical trials often involve complex, multidimensional treatments.
• Estimating minimum effective doses (MED) for such treatments presents statistical challenges, particularly with partially ordered outcomes.

Purpose of the Study:

• To develop a novel statistical method for estimating MED in multidimensional dose-finding trials.
• To address the challenge of partially ordered outcome data in identifying optimal treatment combinations.

Main Methods:

• Proposed an estimation method maximizing a weighted product of posterior gains to circumvent partial ordering constraints.
• Introduced an asymmetric gain function indexed by a decision parameter $$ for balancing true positive and true negative decisions.
• Developed an adaptive $$-tapering algorithm to enhance the identification of effective doses.

Main Results:

• Simulation studies demonstrated that asymmetric gain functions are critical for controlling false discoveries.
• $$-tapering significantly increased the true positive rate (achieving ~90%) compared to nonadaptive designs (~68%) with a comparable false discovery rate.
• The proposed adaptive method showed consistent high true positive rates across various scenarios.

Conclusions:

• The novel estimation method with an asymmetric gain function and adaptive $$-tapering is effective for identifying minimum effective doses in multidimensional treatments.
• This approach enhances the accuracy of dose-finding trials by improving true positive rates while maintaining low false discovery rates.
• The method is particularly relevant for behavioral intervention trials aiming to optimize treatment parameters like sedentary break frequency and duration for glucose level reduction.