Biostatistics 304. Cluster analysis.

In Cluster analysis, we seek to identify the “natural” structure of groups based on a multivariate profile, if it exists, which both minimises the within-group variation and maximises the between-group variation. The objective is to perform data reduction into manageable bite-sizes which could be used in further analysis or developing hypothesis concerning the nature of the data. It is exploratory, descriptive and non-inferential. This technique will always create clusters, be it right or wrong. The solutions are not unique since they are dependent on the variables used and how cluster membership is being defined. There are no essential assumptions required for its use except that there must be some regard to theoretical/conceptual rationale upon which the variables are selected. For simplicity, we shall use 10 subjects to demonstrate how cluster analysis works. We are interested to group these 10 subjects into complianceon-medication-taking (for example) subgroups basing on four biomarkers, and later to use the clusters to do further analysis – say, to profile compliant vs non-compliant subjects. The descriptives are given in Table I, with higher values being indicative of better compliance.