Follow NEMC
 |
 |
 |
The Other Half of Your Lab Result
Oral Presentation
Prepared by B. Vining, Y. Tondeur, J. Hart
SGS Analytical Perspectives, 5500 Business Drive, Wilmington, NC, 28405
Contact Information: bryan.vining@sgs.com; 910-794-1613
ABSTRACT
Consumers of laboratory data – facilities being tested, regulators, risk assessors, and others – often are quite content to receive a single, simple number as the result of the lab’s efforts. Based on this number, one or more individuals will make decisions about human health. Perhaps a facility will be forced to stop operations. Conversely, an area previously deemed unsafe for human habitation may be made into a playground. These decisions frequently occur based on a number that is missing its complete context – the measurement uncertainty. Without this key attribute, the data is incomplete and potentially misleading – especially in cases of significant error.
Many publications exist regarding the estimation of measurement uncertainty, advocating many different means for determining it. Most of these methods proceed from backwards-looking approaches. For example, one may perform a retrospective analysis of proficiency testing data to obtain a static uncertainty interval to apply to future measurements. Not only must one be careful to use appropriate statistics in such cases, but these methods inherently fail to inform the data users about the impact of any changes in the analytical system from the state in which such data were acquired to the state in which the data for their samples. For instance, if an instrument exhibits unusually poor response, the uncertainty may change. We will demonstrate how a system that estimates uncertainty for each measurement – that is, where and when it is needed – works. We will briefly delve into applications for such a system and the value it brings to data users. Most importantly, we will show that uncertainty is not a constant and approaches that respect this fact are needed.
Oral Presentation
Prepared by B. Vining, Y. Tondeur, J. Hart
SGS Analytical Perspectives, 5500 Business Drive, Wilmington, NC, 28405
Contact Information: bryan.vining@sgs.com; 910-794-1613
ABSTRACT
Consumers of laboratory data – facilities being tested, regulators, risk assessors, and others – often are quite content to receive a single, simple number as the result of the lab’s efforts. Based on this number, one or more individuals will make decisions about human health. Perhaps a facility will be forced to stop operations. Conversely, an area previously deemed unsafe for human habitation may be made into a playground. These decisions frequently occur based on a number that is missing its complete context – the measurement uncertainty. Without this key attribute, the data is incomplete and potentially misleading – especially in cases of significant error.
Many publications exist regarding the estimation of measurement uncertainty, advocating many different means for determining it. Most of these methods proceed from backwards-looking approaches. For example, one may perform a retrospective analysis of proficiency testing data to obtain a static uncertainty interval to apply to future measurements. Not only must one be careful to use appropriate statistics in such cases, but these methods inherently fail to inform the data users about the impact of any changes in the analytical system from the state in which such data were acquired to the state in which the data for their samples. For instance, if an instrument exhibits unusually poor response, the uncertainty may change. We will demonstrate how a system that estimates uncertainty for each measurement – that is, where and when it is needed – works. We will briefly delve into applications for such a system and the value it brings to data users. Most importantly, we will show that uncertainty is not a constant and approaches that respect this fact are needed.