Sample Size Determination for Longitudinal Binary Data
*Kush Kapur, Hines VA Hospital 

Keywords:

We consider the problem of sample size determination for testing the group by time interaction parameter for two-level longitudinal mixed-effects logistic regression models. We derive a closed form solution for determining the necessary sample size accounting for attrition rates across time in order to achieve a pre-specified power (80%, 85% or 90%) while strictly maintaining the Type I error rate (e.g. 5%). The method has been applied to sample size and power calculation in a Veteran Affairs CCTA study titled "Rivastigmine patch in Veterans with Cognitive Impairment Following Trauma Brain Injury". The presented approach lead to reduction in sample size requirement compared to other available methods.