Imputation Strategies When a Continuous Outcome Is to Be Dichotomized for Responder Analysis: a Simulation Study

• Activity Number: 299 - Estimands and Imputations Methods
• Type: Contributed
• Date/Time: Tuesday, July 30, 2019 : 8:30 AM to 10:20 AM
• Sponsor: Biopharmaceutical Section
• Author(s): Lysbeth Floden* and Melanie Bell
• Companies: University of Arizona and University of Arizona
• Keywords: Responder analysis; Clinical trials; Multiple imputation; Missing data; Patient reported outcomes; Missing at random

Abstract

In many clinical trials continuous outcomes are dichotomized to compare proportions of responders. A common approach is to impute missing outcomes as non-responders, despite known biases. In this study we compared multiple imputation (MI) approaches to imputing as non-response in responder analysis. Methods. We simulated a two-arm trial with a continuous outcome at four time points. We omitted data using missing at random mechanisms, and imputed by replacing as non-responder, imputing before dichotomizing; and imputing binary response. We assessed bias, power, and type 1 error. We applied these methods to a clinical trial. Results. MI performed better than non-response imputation in terms of bias and type 1 error. When 30% of responses were missing, bias was less than 6% for all MI scenarios. Non-response imputation resulted in both under- and overestimates. In the example, non-response imputation estimated a smaller difference in response than MI approaches. MI performed better than imputing missing observations as non-responders. Multiply imputing the continuous outcome variable prior to dichotomizing was slightly less biased than multiply imputing the responder status.