On the interpretation of microbial clusters in periodontal disease

Cluster analysis is sometimes used in periodontal research to describe patterns of microbial identifications observed among cases (subjects or sites). These patterns are assumed to define subsets of cases that are meaningful, in the sense of defining periodontal subgroups that have reliable differences in some aspect of their disease. However, cluster analysis is an exploratory, first-stage technique that does not incorporate the machinery that allows one to evaluate the statistical significance and reliability of the patterns observed. Problems inherent in drawing inferences from cluster analysis are discussed and, as one methodological example of how the statistical significance of clustering might be determined, a randomization-based test for binary scaled microbial identifications is described. The introduction of cluster analysis to the study of periodontal microbiology was important and innovative, but future work needs to incorporate validation by either statistical means or independent replication. The present rarity of such confirmatory methods in the literature suggests caution in accepting specific cluster-based hypotheses about patient-microbial patterns.