In terms of clustering, you could do a principal components analysis  (e.g. http://europepmc.org/abstract/med/23091073) or perhaps a latent class analysis (http://www.biomedcentral.com/1471-2318/12/64). 

The problem is how to define clustering of risk factors - you can use a random combination of a sub-sample of risk factors and repeat it many time to see if there are the robust patterns. As a rule, the risk factors are not independent, even if sometimes they can be regarded as being "independent". I do not recall anything very valuable that has been published about multimorbidity and clusters Iof course I might miss something) except some works of Barabasi team about networks. One paper is Hidalgo et al in Plos Computational Biology "A dynamic network approach for the study of human phenotypes".  I am not aware of any applications of this approach to multimorbidities; I tried to take a look at that in a broader context of multiple symptoms (except morbidities) but cannot give you a reference yet. In general, Based on my personal experience, I am in favor of the network approach rather than "conventional" clustering algorithms (I have never seen anything robust and generalizable outside of one dataset where such algorithms were applied).

Adding my bit to that of the others, I would also give canonical correlation analysis (CCA) a try. It is a very informative data exploration method similar to PCA to simultaneously look at the pattern of cross-variances between (and within) two sets of variables, like in your case could be the morbidities (set 1) and the risk factors (set 2). The CCA will find the linear combination of these variables that will have maximum correlation with each other. This way you may see at once which are the correlations between the the risk factors and the morbidities, but also the correlations among the risk factors and among the morbidities themselves.