RESUMO
The statistical methods and parameters commonly used to define bacterial susceptibility to antibiotics in vitro such as MIC(50), linear regression or others, usually lead to a considerable loss of information: they do not take into account the heterogeneity of the bacterial population. In contrast, multivariate data analyses are more adapted to the description of biological systems. In this way, a population of a given bacterial species can be separated into homogenous classes corresponding to the different sensitivity and resistance phenotypes. The applications of this mathematical approach include: (i) a new model for more relevant interpretation of antimicrobial susceptibility test results; (ii) numerical estimation of breakpoints having a known risk; (iii) calibration of a technique relative to a reference technique; (iv) detection of strains with new phenotypes; (v) in vitro evaluation of the activity of new compounds.