Re-Examination of MMRM and ANCOVA.LOCF in an Acute Schizophrenia Trial

• Activity Number: 244
• Type: Contributed
• Date/Time: Monday, July 30, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Yangchun Du*+ and Asli Memisoglu and Marc L de Somer
• Companies: Alkermes, Inc. and Alkermes, Inc. and Alkermes, Inc.
• Keywords: schizophrenia ; missing data ; MMRM ; ANCOVA ; informative missing

Abstract

In longitudinal clinical trials with missing data, recent studies reported that MMRM is generally considered superior to ANCOVA.LOCF for type 1 error rate control, accuracy and precision of effect estimates. These studies explored cases with drop-out rates < 50%, and informative missing data < 30%. However, a typical acute schizophrenia trial has a drop-out rate in excess of 50%, with differential drop-out between treatment and placebo, and a high proportion of informative missing data.

We have examined the behavior of ANCOVA.LOCF and MMRM under these circumstances through comprehensive simulation studies.

Our results reveal that MMRM produced an approximately unbiased estimate and controlled the type 1 error when drop-outs were missing at random and non-differential. MMRM produced significant bias when the drop-out rate difference was greater than 10%, and the amount of bias depended on the proportion of informative missingness. In contrast, ANCOVA.LOCF generally underestimated the effect, inflated standard error, and resulted in conservative estimation and inference.