Optimised point estimators for multi-stage single-arm phase II oncology trials

Area of Science:

• Biostatistics
• Clinical Trial Methodology
• Statistical Estimation

Background:

• The uniform minimum variance unbiased estimator (UMVUE) is a standard for unbiased estimation in multi-stage single-arm trials with dichotomous outcomes.
• However, the UMVUE often exhibits a large residual mean squared error (RMSE), indicating substantial estimation inaccuracy.
• Existing methods struggle to balance bias and variance effectively for various response rates.

Purpose of the Study:

• To develop an optimization approach for creating statistical estimators with reduced RMSE in clinical trials.
• To identify estimators that achieve low bias alongside minimized RMSE across a range of response rates.
• To provide a more accurate estimation strategy than the traditional UMVUE.

Main Methods:

• Formulated an optimization problem to systematically search for improved estimators.
• Investigated the performance of proposed estimators across diverse response rate scenarios.
• Analyzed the trade-off between bias and RMSE for the developed estimators.

Main Results:

• The optimization approach successfully identified estimators with substantially reduced RMSE compared to the UMVUE.
• Careful selection of optimization parameters led to estimators with minimal introduction of appreciable bias.
• Demonstrated improved estimation accuracy for various response rates commonly encountered in clinical trials.

Conclusions:

• The proposed optimization method offers a viable strategy for enhancing statistical estimation in clinical trials.
• This approach provides estimators that effectively balance bias and RMSE, outperforming the UMVUE.
• The findings suggest a practical improvement for analyzing dichotomous outcomes in multi-stage single-arm trials.