Statistical analysis of the effects of trial, reader, and replicates on MIC determination for cefoxitin

Abstract

A pilot study was designed to estimate the variance components in the determination of the MIC of cefoxitin for isolates of the Bacteroides fragilis group. Twenty different organisms were tested, and replicate, trial, and reader variabilities were examined. When the total-variance component was used, if the true MIC was 16 micrograms/ml, then the chance that the observed MIC was between 8 and 32 micrograms/ml, inclusive, was 95%. For all analyses, the isolate (P = 0.0001) and reader (P less than 0.03) effects were significant. The probability of specific MIC observations for various true MICs (over the range of 16 to 32 micrograms/ml at 4-micrograms/ml increments) was calculated. For true MICs of 20, 24, and 28 micrograms/ml, the probabilities of observing an MIC of 16 or 32 micrograms/ml (inclusive) were 86, 75, and 62%, respectively. An upward bias was shown to exist in addition to sources of sizeable variation. The recommendation stemming from recognition of this inherent variability is that ranges of percent susceptibility at various concentrations be included in reports of in vitro susceptibility studies.