Bayesian variable selection for longitudinal substance abuse treatment data subject to informative censoring