To determine the susceptibility of an unknown pathogen to a specific drug, hospital laboratories perform either a drug dilution or disk diffusion test. Test/Drug specific breakpoints then classify the pathogen as either susceptible, intermediate or resistant to the drug. Because only one of these tests will be used in practice, it is imperative to have the two tests give similar classification results. It is the responsibility of subcommittees of the FDA and NCCLS to determine comparable breakpoints. Since the drug dilution test involves concentrations of the drug, its breakpoints are based largely on the pharmacokinetics and pharmacodynamics of the drug. Currently, disk diffusion breakpoints are determined using a modified version of the error-rate bounded method. This involves the creation of a scatterplot of test results for a wide range of pathogens and finding breakpoints which minimize the number of classification discrepancies. While this procedure is very fast and simple to implement, it does not adequately take into account the inherent variability of each test, the drug-specific relationship between the two tests, nor the underlying distribution of pathogens. As a result, the procedure's choice of breakpoints is very sample dependent. For this project, an alternative method of breakpoint determination will be developed which uses the same scatterplot of test results but explicitly account for the various factors of uncertainty using a hierarchical model. Probabilities from this model, which do not depend on the distribution of isolates, will then be used to determine appropriate breakpoints. While this procedure will be more time consuming to implement (particularly the inference component), it is believed that by accounting for the inherent test variabilities and underlying distribution of pathogens, much of the sample dependency will be eliminated giving more consistent and interpretable results.
| Liu, Yali; Craig, Bruce A (2006) Incorporating time-dependent covariates in survival analysis using the LVAR method. Stat Med 25:1729-40 |
| Annis, David H; Craig, Bruce A (2005) Statistical properties and inference of the antimicrobial MIC test. Stat Med 24:3631-44 |
