Comparison of Smoke Yield Data Collected from Different Laboratories

• Main Authors: Teillet B, Cahours X, Verron T, Colard S, Purkis SW
• Format: Article
• Language: English
• Published: Sciendo 2013-12-01
• Series: Beiträge zur Tabakforschung International
• Online Access: https://doi.org/10.2478/cttr-2013-0943

In the context of increasing tobacco product regulations, more requirements are observed for the reporting of smoke constituent yield data and its variability e.g., US Food and Drug Administration (FDA). The objective of this work was to evaluate the relevance of the short term standard deviation to describe the variability of measurements using the dataset of the CORESTA 2006 Joint Experiment which included a number of cigarette smoke constituents more recently identified by FDA for reporting. Their testing protocol required the analysis of Kentucky Reference cigarettes 2R4F and 1R5F performing five replicates run over consecutive days, repeated during three different time periods. This dataset provided access to different sources of smoke yield variability across measurements: short term and medium term within-laboratory variability and among-laboratory variability. For each reference cigarette, analysis of variance on one factor (laboratory) combined with the Newman-Keuls multiple range test was performed to compare data generated across laboratories. Results showed that the expression of yield variability as an individual standard deviation (describing repeatability) gives erroneous conclusions due to the major contribution of amonglaboratory variability not being taken into account. The different sources of variability can be taken into account in the comparison using the critical difference, as described in the ISO Standard 5725 part 6. This paper shows the importance of having i) the appropriate statistical methods to compare results from different laboratories in order to avoid erroneous conclusions, and ii) validated and standardized methods with known precision across laboratories. Moreover, it was demonstrated that the number of replicates had only a minor effect on product comparison on the basis of the critical difference as a function of repeatability and reproducibility of the methods.