Cluster analysis of genetic and epidemiological data in molecular epidemiology.

Current molecular epidemiological studies of complex diseases include a large number of genetic and epidemiological variables. Clustering approaches are a useful tool to detect patterns in data sets and generate hypothesis regarding potential relationships in complex data situations. In this article similarity coefficients are presented for a hierarchical cluster analysis of single-nucleotide polymorphisms (SNPs) and epidemiological data to gain insight into the relationship of variables and detect potential differences between diseased and control individuals in case-control studies. This approach was applied to two subsets of data from the GENICA study of sporadic breast cancer, a molecular epidemiological population-based case-control study conducted in the greater Bonn region between 2000 and 2004. Separate cluster analyses for cases and controls using flexible matching coefficients for SNPs, Pearson's corrected coefficient of contingency for categorical epidemiological variables, and Spearman's correlation coefficient for quantitative epidemiological variables as measures of similarity revealed small subgroups of SNPs usually of the same gene, as well as clusters of genetic and of epidemiological variables with minor differences between cases and controls. In addition to recent and well-known findings, the joint cluster analysis of SNPs and epidemiological variables provides further insight into the relationship of these variables.