Cluster analysis of antibiotic susceptibility patterns of clinical isolates as a tool in nosocomial infection surveillance

Abstract

Hospital infections represent a major epidemiological problem. The first step in the detection of nosocomial infections consists in assessing the probability that two or more isolates from different patients are similar or different. Many methods are available for typing purposes. Among these, antibiotic susceptibility patterns do not need extra cost or extra work and are available "on line" every moment they are needed. A mathematical technique of elaboration is proposed for disk zone sizes, in order to assess the probability of two or more clinical isolates to be the same strain. Antibiograms performed according to Kirby-Bauer are evaluated detecting zone sizes by a computer controlled device and then submitted to cluster analysis. Similarity of strains is reported in a dendrogram, in which strains are successively fused. Strains that share a common susceptibility pattern are considered a "cluster". At last, epidemiological maps are constructed for each group of strains, in which all the isolates are reported, ordered for patients, plotted on the day the specimen was collected, drawn in a different shape according to the source of specimen, and shadowed by the pattern of its cluster. This method of reporting data directly allows to detect cross infections among patients and can be used as a first typing step before other more expensive procedures.