Unsuitability of MIC50, MIC90 for comparison of antibiotic potency
Abstract number: P2059
Objective: The potency of antibiotics is commonly compared by ranking their MIC50 or MIC90. This method is plainly unsuitable when there are distinct resistant subpopulations, but may appear attractive for unimodal MIC distributions. Its performance was investigated by simulation.
Methods: Simulations, repeated 1000 times, compared two antibiotics, A and B, each tested on all of 20, 50, 100 or 500 isolates. Simulation and analysis used the log2 scale where conventional doubling-dilution MICs (0.25, 0.5, 1, 2 mg/L, etc.) become integers (-2, -1, 0, 1, etc.). The MIC peak for A was set at 0, 0.25, 0.5 or 0.75 i.e. exactly at, or at various levels between, conventional MICs. Intrinsic variation between isolates and experimental variation were both modelled as continuous normal distributions, giving continuous 'underlying' MICs that were rounded up to integers when 'measured'. Experimental variation had an SD of 0.3, 0.4 or 0.6 (i.e. about 99%, 94% or 79% of replicate MICs within±1 dilution). The true relative potency of A and B was constant in all isolates, differing by 0, 0.25, 0.5, 1 or 2 dilutions, but subject to experimental variation.
Results: When both drugs had equal potency, their MIC50 and MIC90 often showed apparent differences in over 40% of simulations in some cases, even when experimental variation was low. Conversely, simple comparison of MIC50 or MIC90 could consistently fail to detect differences of 0.25 or 0.5 doubling dilutions, even with the largest sample sizes and least experimental variation, if both fell within the same doubling dilution band. The matched-pairs Wilcoxon signed-ranks test had a false-positive error rate close to an acceptable 5% level in all cases. Its power fell with increasing experimental variation (e.g. 95%, 85%, and 61% for a 0.5 dilution difference with N = 20), but increased with sample size and was unaffected by the exact position of the MICs relative to doubling dilution bands.
Conclusion: Simple comparison of MIC50 or MIC90 is unsuitable for the ranking of antibiotic potency: it performs erratically, dependent on the exact location of MICs and regardless of large sample size, and often indicates spurious differences or fails to detect real differences. The Wilcoxon signed-ranks test is safer and more powerful, and a description of paired MIC differences is more informative. More detailed understanding of the variation inherent in MIC methodology is needed for proper assessment of methods to compare MICs.
|Session name:||18th European Congress of Clinical Microbiology and Infectious Diseases|
|Back to top|