We propose a multistate modeling approach to describe the observed evolution of patients diagnosed with non-muscle-invasive bladder cancer. On the basis of data from the Spanish Bladder Cancer/EPICURO study, we adjust a multistate model taking into account the disease-related events of interest (recurrence, progression, and disease-related deaths) as well as competing deaths due to other causes. We then develop a dynamic predictive process for bladder cancer progression, which allows the risk of a patient to be updated whenever new information of his or her evolution is available. By using specific measures of prospective accuracy in the presence of competing risks, the proposed dynamic model has shown to improve prediction accuracy and provides a more personalized management of bladder patients.
Copyright © 2011 John Wiley & Sons, Ltd.