|
Abstract:
|
The handling of clinical trial data after use of rescue therapy warrants careful statistical considerations. Our simulation work is conducted in the context of clinical development planning for a devastating rare disease where there is an existing therapy. The primary endpoint is a neurological functional score collected by visits, and data after taking the existing therapy as rescue therapy are treated as missing. Through a simulation study, we compared several commonly used methods, including Mixed Model for Repeated Measure (MMRM), constrained longitudinal data analysis (cLDA), multiple imputation (MI), trimmed mean, and copy increment to reference (CIR), etc. Different simulation setups are considered, including different designs (e.g. single arm vs. randomized controlled trial), missing proportions, and missing mechanism (MAR vs. MNAR). The simulation results help us understand the limitation and performance of these methods, which is critical to trial design.
|
