BAYESIAN METHODS FOR JOINT MODELING OF LONGITUDINAL AND SURVIVAL DATA WITH APPLICATIONS TO CANCER VACCINE TRIALS

Vaccines have received a great deal of attention recently as potential therapies in cancer clinical trials. One reason for this is that they are much less toxic than chemotherapies and potentially less expensive. However, little is currently known about the biologic activity of vaccines and whether they are associated with clinical outcome. The antibody immune measures IgG and IgM have been proposed as potential useful measures in melanoma clinical trials because of their observed association with clinical outcome in pilot studies. To better understand the role of the IgG and IgM antibodies for a particular vaccine, we examine a case study in melanoma and investigate the association between clinical outcome and an individual’s antibody (IgG and IgM titers) history over time. The Cox proportional hazards model is used to study the relationship between the antibody titers as a time varying covariate and survival. We develop a Bayesian joint model for multivariate longitudinal and survival data and give its biologic motivation. Various scientific features of the model are discussed and interpreted. In addition, we present a model assessment tool called the multivariate L measure that allows us to formally compare different models. A detailed analysis of a recent phase II melanoma vaccine clinical trial conducted by the Eastern Cooperative Oncology Group is presented