The term “networking” is often used to describe the building up of social contacts or joining distant parts of the world with communication cables. The defining characteristic of these processes is the lack of centralized command, that is, it is difficult to determine beforehand what links will be formed and when. On the other hand, complex network theory has revealed shared universal properties between most real-world networks, which has prompted the widespread enthusiasm about networks as tools to understand complex phenomena.
To jump from social networks to clinical data analysis is not obvious at first glance. However, the two problems are, in fact, similar. For instance, a mobile communications register is an incomplete sample of the total communicative activity between humans, and special interest is on the community characteristics (spontanious grouping of people) within the network. In a clinical study, the interest is on the measured variables (risk factors) and how they can be grouped and whether they reflect underlying metabolic pathways. In the former case, the people are considered nodes and the links quantified by frequency of calls; in the latter case, the variables are regarded as nodes and the links are quantified by the strength of statistical association within the study population. The topology of the network is the target of investigation in both cases.
Katiska is a light-weight tool for calculating the correlation structure within a typical clinical study. It is designed for situations with a moderate number of variables and a large set of patients. It may also be instructive to visit the Himmeli web page for more details on the visualization algortihm.
