|M.Sc Student||Oren Hila|
|Subject||Estimation of the Uncertainty in Chemical Testing with|
Assistance on Horwitz's Predictive Model
|Department||Department of Quality Assurance and Reliability||Supervisors||DR. Haim Hacham|
|PROFESSOR EMERITUS Amos Notea|
|Full Thesis text - in Hebrew|
Standards of the International Organization for Standardization (ISO) enable the customer to have better confidence in the supplier. Responsible decisions as to whether or not a reported result exceeds a regulatory limit can be made only with full knowledge of the measurement uncertainty. Accordingly, Laboratories are expected to have and apply estimation procedures for the uncertainty measurement in all their relevant test methods, in parallel with reliance on previous experience.
Thus, the laboratory's finding is supposed to be in the following form: Result = Value ± Uncertainty
Over the last years, guides were published, proposing estimation methodologies for the measurement uncertainty: for example, ISO GUM-Guide to the Expression of Uncertainty in Measurement (1993) and the EUROCHEM interpretation of the latter in the field of analytical chemistry (1995). According to these guides, the estimation process involves multiple tasks: identification, quantification and combination of all potential sources of uncertainty. Difficulties arise in the case of complex chemical tests such as high-pressure liquid chromatography (HPLC), in which the number of possible components likely to contribute to the uncertainty is almost infinite.
In these circumstances, concerns have been expressed about the time and resource requirements, especially for laboratories with a wide range of tests and those with low resources. The present final paper proposes a model which resolves this difficulty with a minimal effort. In this model, based on data obtained from the regional Public Health Laboratory in Haifa and published data, the calculated uncertainty estimate is compared against the maximum estimate predicted by the generic Horwitz model. The chemical test chosen for verifying the model is the determination of histamine in frozen fish products by HPLC.
The estimation procedure was simplified by classifying the uncertainty sources in four "families" in terms of precision, namely: method recovery, calibration, homogeneity, and intra-laboratory reproducibility.
The calculated combined and expanded uncertainties for a total of 200 ppm histamine resulted in a value of 200 ± 27 ppm, as against 200 ± 29 ppm which is predicted by the Horwitz model, i.e. below the upper uncertainty limit. Accordingly, the laboratory is entitled to declare the Horwitz estimate (which is stricter) and, if it so desires, declare a lower value later subject to approval by the accreditation authorities. If the proposed model yields a higher estimate than the Horwitz prediction, the estimate should be re-examined more thoroughly. This proposed model might although be applicable to a broad range of tests carried out in testing laboratories.