Comparison of Parametric and Non-parametric Approaches for Accuracy of Quantitative Microbiological Methods

Quantitative microbiological methods are aimed at counting the number of microorganisms in a sample. They are extremely important in the pharmaceutical industry to ensure drug safety: indeed, bacteria and toxins produced by microorganisms could contaminate medicines, which may harm humans if the contamination remains undetected. Many new technologies are being developed to measure the microbial content in a sample: they are generally called rapid methods because they provide much quicker results than compendial methods, which are standardized methods and need up to 14 days to produce a result. Rapid methods need to be validated before being practically used and accuracy is one of the parameters which needs to be evaluated during validation. Essentially, the accuracy of the rapid method is evaluated by comparing its expected measurement to the expected measurement of a compendial method. This Master thesis consists of an in-depth comparison of statistical methods to assess the accuracy of a microbiological method. One approach is parametric, since it is based on the estimation of two generalized linear models, while the other one is non-parametric. The former is referred to as model-based approach and the latter is referred to as non model-based approach. A simulation study is performed to compare the performances of the two approaches in terms of correct assessment of accuracy of the rapid method. The results show that it is not possible to definitely declare that one of the two approaches is preferable to the other one in any situations, but some interesting patterns can be derived in the performances of the two approaches. Finally, the design used to estimate the linear models in the model-based approach is optimized to improve the performance of this approach. Some interesting conclusions can be derived from this analysis. However, many questions remain unresolved and could be the basis for future work, especially with respect to the use of the model-based approach and the optimal design.