Statistical Trade-Off of Using Rescue Medications in Psychiatric Clinical Trials

• Activity Number: 554
• Type: Topic Contributed
• Date/Time: Wednesday, August 6, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Mental Health Statistics Section
• Author(s): Zhibao Mi*+ and Julia Lin and John H. Krystal and Karen M. Jones and Robert A. Rosenheck and Mei Chiung Shih and Kousick Kousick Biswas and Joseph F. Collins and Robert A. Lew and Ying Lu
• Companies: Department of Veteran Affairs and Palo Alto VA CSPCC and VA Connecticut Healthcare Systems and VA Cooperative Studies Program Coordinating Center and Yale School of Medicine and Stanford University School of Medicine and VA Cooperative Studies Program Coordinating Center and VA Cooperative Studies Program Coordinating Center and VA Cooperative Studies Program Coordinating Center and Palo Alto VA CSPCC/Stanford University
• Keywords: Missing data ; Rescue medication ; Dropout ; Psychiatric trial

Abstract

Missing data could seriously compromise the statistical inferences of clinical trials. Data missing in psychiatric trials are primarily caused by participant dropouts which are often due to worsening symptoms and ineffective treatments. Dropout can be mitigated by allowing use of rescue medications for participants who are not responding to the study treatment. Little et al recommended eight ways for preventing missing data through the design of clinical trials. One of the recommendations is to allow the use of rescue medications that are designated as components of a treatment regimen in the study protocol. Though this design feature can reduce dropout and limit missing data, it may introduce bias and additional variance. Here we use statistical modeling to evaluate when a rescue medication should be introduced for the purpose of participant retention and missing data reduction and the trade-off between maximizing analyzable sample size and noise reduction. Data from a recent VA randomized psychiatric trial are used to illustrate the potential impact of rescue medication use on the inference and the possible bias introduced to the study.