An approach to integrated data assessment in a proficiency test on the enumeration of Escherichia coli

Abstract

To develop appropriate statistical approaches to plan and evaluate proficiency tests for the enumeration of Escherichia coli, addressing, in particular, a possible but frequently unavoidable lack of test sample homogeneity. Each of 50 laboratories analysed two samples of a stabilized suspension of E. coli in duplicate, using various media, inoculation methods, and incubation times and conditions. In parallel, the E. coli suspension was tested by the organiser for homogeneity and stability. Escherichia coli counts followed a log-normal distribution. After eliminating, by Youden analysis, two data sets that were considered outliers and eight data sets for underperformance of the laboratories (substantial lack of repeatability), the standard deviation of the mean was about 0·06 log(10) units. There was no evidence of bimodality of the data. Lack of homogeneity of distribution of bacteria had a strong effect on measurement uncertainty, in addition to laboratory bias and method repeatability. The homogeneity decreases during storage of the individual test vials; this effect could be modelled by the known kinetics of inactivation of micro-organisms. The results were confirmed by Monte Carlo simulations. By a tailored analysis of proficiency testing data, it is possible to distinguish the effect of lack of homogeneity, laboratory bias and method repeatability, on the measurement uncertainty. A statistic tool is provided to solve problems related to lack of stability of microbiological test material and to separate the effects of sample inhomogeneity from the performance of the individual laboratory.