Estimation after subpopulation selection in adaptive seamless trials

Area of Science:

• Biostatistics
• Clinical Trial Design
• Pharmacoeconomics

Background:

• Developing new therapies often involves testing treatments in the full patient population or specific subpopulations.
• Traditional approaches use separate trials for population selection and confirmation, which can be inefficient.
• Adaptive seamless designs (ASDs) offer a more efficient two-stage approach, using stage 1 data for population selection and stage 2 for confirmation.

Purpose of the Study:

• To investigate the statistical challenges, specifically selection bias, introduced by using stage 1 data for both population selection and confirmatory analysis in ASDs.
• To describe the extent of bias in estimators that do not account for multiple hypotheses and population selection.
• To derive and examine the properties of conditionally unbiased estimators for point estimation in ASDs.

Main Methods:

• Focus on point estimation within the context of two-stage adaptive seamless designs.
• Analyze bias in estimators that ignore the selection process based on stage 1 data.
• Derive conditionally unbiased estimators and evaluate their mean squared errors under various scenarios.

Main Results:

• Characterization of selection bias in ASDs when stage 1 data influences population selection for stage 2.
• Development of novel estimators designed to mitigate bias arising from the adaptive selection process.
• Evaluation of the performance of these new estimators compared to traditional ones in terms of mean squared error.

Conclusions:

• ASDs are efficient but introduce statistical complexities, particularly selection bias in point estimation.
• The derived conditionally unbiased estimators provide a statistically sound method for inference in ASDs.
• This research contributes to more reliable and accurate evaluation of new therapies using adaptive trial designs.