Simulated Clinical Trial Data with Nonlinear, Nonignorable Dropout
Samantha MaWhinney (UCD), Jeri Forster (UCD)

S-PLUS software (Insightful Corp., Seattle WA)
Dropout is common in clinical trials and other longitudinal studies. When the reason for dropout is associated with the outcome of interest, nonignorable missing data are present. Ignoring this phenomenon in the analysis will lead to biased results. This graph contains nine realizations of simulated clinical trial outcome data. The x- and y-axes represent time on study and the longitudinal dropout-varying outcome. Each realization assumes 100 subjects with a nonlinear dropout-varying slope of ­exp(-4u). For each subject the dropout time, u, is randomly generated using a beta-binomial random number generator. Possible values of u include 0, 1/15, 2/15, …, 1. Random subject effects and within subject variability are also incorporated to provide realistic data to be used for comparing the performance of new and existing statistical methods.