Mixed Effect Model for Missing Not at Random in Xenograft Tumor Growth Assays

• Activity Number: 438
• Type: Contributed
• Date/Time: Tuesday, August 6, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Health Policy Statistics Section
• Author(s): Xiaoli Shirley Glasgow*+ and George Naumov and Kuenhi Tsai
• Companies: Merck & Co, Inc and Merck and Merck
• Keywords: missing not at random ; xenograft assay ; mixed effect tumor growth model

Abstract

Xenograft tumor assays are in vivo methods of screening investigative cancer drugs, in which human tumor tissue or cells are xenografted into mice or rats followed by tumor treatment regimens. The tumor growth is then subsequently monitored for the effects of treatments. Often some mice/rats died or are sacrificed prior to the end of experiment. This causes missing data not at random (MNAR). Inappropriately handling MNAR, such as using standard statistics or software, may lead to biased estimates and incorrect experimental conclusions. Common practices for MNAR are selection model and pattern-mixture model mainly designed for clinical trials or surveys. Very few applications were published in preclinical area. We propose a new imputation approach by using mixed effect tumor growth model. This new approach is specially designed for in-vivo xenograft tumor growth assays, in which the missing data mechanism is captured by a mixed tumor growth model. It simplifies the analysis greatly compare to the selection model and pattern-mixed model. The simulation studies demonstrate the strength of this new approach.