In clinical research, we are often interested in assessing how a biomarker changes with time, and whether it could be used as a surrogate marker when evaluating the efficacy of a new drug. However, when the longitudinal marker is correlated with survival, linear mixed models for longitudinal data may be inappropriate. By contrast, it may be possible to recover information from the so-called informative censoring by modelling both the longitudinal information and the survival process. The objective of this work is to jointly model longitudinal and survival data to assess surrogacy. Two competitive modelling strategies were used, either a multistate model summarizing the course of longitudinal data and occurrence of disease progression or death, or a joint longitudinal-survival model. We present both analyses based on a case study from two randomized clinical trials that enrolled patients with stage A chronic lymphocytic leukaemia (CLL) in order to obtain further insights into these different approaches.
Copyright (c) 2007 John Wiley & Sons, Ltd.