|
Abstract:
|
Incomplete data analysis continues to be one of major issues for non-inferiority clinical trials. Due to the steadily increasing use of non-inferiority study design, we believe this topic deserves immediate attention. We evaluate different missing data procedures together with complete data methods for incomplete non-inferiority trials focusing on a difference in binomial proportions. Here we present results from a comprehensive simulation study. We consider a fixed margin approach and assess the following methods for construction of confidence intervals: Wald, Farrington-Manning and Newcombe. First, we compare performance of these methods for fully observed data under various study scenarios. Then, each method is assessed for partially observed data employing different incomplete data analysis strategies, such as complete case analysis, best and worst case scenario imputation, and multiple imputation. We show that multiple imputation outperforms other incomplete data analysis strategies we considered. We conclude this simulation study with recommendations regarding incomplete data analysis for non-inferiority clinical trials which evaluate difference in binomial proportions.
|
