132 – Analysis of Missing Data
Multiple Imputation for Cytokine/Metabolite Assay Data with Missing Data and Values Below Detectable Limits of Quantification
Ying Guo
Merck
Saijuan Zhang
Merck
Many clinical trials data are complicated by the existence of fully missing values or left-censored values known to lie below detection limits, due to biological reasons or assay technical limitations. A conventional practice is to use the actual Lower Limit of Quantification (LLOQ) into the missing value. In this article, we describe a multiple imputation (MI) method for multiply imputing the missing and left-censored values of cytokines and metabolites to understand the potential treatment effects on these biomarkers. A key advantage of multiple imputation is that, once multiple imputed data sets are created, standard analysis methods for complete data can be applied, with imputation uncertainty being addressed by applying MI combining rules. It also provides a convenient approach to limit of quantification issues. We compare the proposed MI method with the existing methods including the conventional substitution analysis and the simple imputation techniques, and demonstrate its superiority in terms of relative bias, efficiency and coverage probability for the 95% confidence interval through simulation studies and a real study which evaluated active treatments on biomarkers to allergen in asthmatics.