Measurement Error in Clinical Trials and sample Size Determination

Abstract

Adaptive clinical trials offer a flexible approach to refining sample sizes during ongoing research, enhancing trial efficiency. This study delves into improving sample size recalculation through resampling techniques, employing measurement error and mixed distribution models. The core inquiry addresses the potency of resampling in enhancing sample size recalculation and evaluates the impact of measurement error and mixed distribution models on clinical trial efficacy. The research employs diverse sample size recalculation strategies—standard simulation, R1, and R2 approaches—where R1 considers the mean and R2 employs both mean and standard deviation as summary locations. These strategies are tested against observed conditional power (OCP), restricted observed conditional power (ROCP), promising zone (PZ), and group sequential design (GSD) on data generated from measurement error and mixed distribution models. The key findings indicate that the R1 approach, capitalizing on mean as a summary location, notably outperforms standard recalculations without resampling, as it mitigates variability in recalculated sample sizes across effect sizes. The OCP exhibits superior performance within the R1 approach compared to ROCP, PZ, and GSD due to enhanced conditional power. However, a tendency to inflate the initial stage’s sample size is observed in the R1 approach, prompting the development of the R2 approach that considers mean and standard deviation. The ROCP in R2 approach demonstrates robust performance across most effect sizes, although GSD retains superiority within R2 approach due to its sample size boundary. Notably, sample size recalculation designs perform worse than R1 for specific effect sizes, attributed to inefficiencies in approaching target sample sizes. In conclusion, resampling-based approaches, particularly R1 and R2, offer improved sample size recalculation over conventional methods. R1 approach excels in minimizing recalculated sample size variability, while R2 approach incorporating both mean and standard deviation, presents a refined alternative. However, chal lenges in precisely approaching target sample sizes under certain conditions indicate avenues for further refinement. This research contributes to the optimization of adaptive clinical trials, enhancing their efficiency and reliability.