Regency EF
An Evaluation of the Trimmed Means Approach in the Context of Randomized Controlled Trials (304030)
*Doug F Arbetter, Veristat
Nicole Martin, Veristat
Keywords: Trimmed Means, Missing Data, Randomized Clinical Trials, Simulation, Multiplicity
Missing data is a concern in clinical study execution and analysis. Missing data leads to biased estimates and results that may not be generalizable, often derailing an investigational product’s chances of regulatory approval without further study. Excluding missing data undermines a study’s power and the ITT paradigm. Imputation methods are dependent on data being missing at random which is often not the case. The Trimmed Means statistical approach (Permutt and Li, 2017) can be employed to accommodate missing data, regardless of reason, and still maintain the ITT principles. Using simulations, we evaluated the operating characteristics of the ANCOVA based approach, compared with a traditional ANCOVA model, in a simulated clinical trial comparing 2 treatment arms to 1 control. At missing data rates of 10% to 30%, by 5% increments, the lower limit of the 95% CI around the power of the traditional ANCOVA analysis never exceeds a difference of 1 from the upper limit of the 95% CI around the power of the fixed trimmed means analysis. Thus, the benefit to the ITT paradigm likely outweighs any negligible loss of power using the trimmed means method compared to the traditional ANCOVA.