Multivariate disease mapping has received considerable attention for the joint spatial study of several diseases. Multivariate disease mapping models estimate the risk of several diseases in specific locations using information from the corresponding spatial neighbors as well as from the rest of diseases. In this way, a greater amount of information is used in the estimation of the risks. Multivariate disease mapping models are computationally slow, therefore, most of the existing literature study at most three or four diseases. Recently, Martinez-Beneito (2013) developed a general framework for multivariate disease mapping able to reproduce many of the previous proposals in the literature. Subsequently, Botella-Rocamora et al. (2015) extended that work, developing a simpler and more convenient form able to handle tens of diseases. Moreover this model can be implemented in Bayesian simulation packages such as WinBUGS. In this work, we present an application of the Botella-Rocamora's methodology to the joint study of several diseases in three Spanish cities. After exploring the results, some limitations of that methodology are evidenced, when applied to mortality data of smaller cities. For this reason, we propose a modification of that methodology which allows solving the problems found and improving risk estimates.