Background:
The European Tobacco Product Directive 2014/40/EU prohibited cigarettes and roll-your-own tobacco having a characterizing flavor, which is defined as "a clearly noticeable smell or taste other than one of tobacco […]". To practically distinguish between products with and without a characterizing flavor, some questions need to be answered. For example, which products should be used as reference? Which sensory method should be used? Should we smell or smoke tobacco? Should we use consumer or expert panels? Which statistical cut-off value should be used to define 'clearly noticeable'? Considering these issues, we developed a sensory method to identify characterizing flavors for regulatory purposes.

Methods:
An expert panel (n=18) was trained to evaluate the smell of twenty tobacco products using self-defined odor attributes. Three types of training sessions were distinguished: attribute training, consensus training, and performance monitoring. Products were tested using Quantitative Descriptive Analysis during six test sessions at Wageningen University (the Netherlands). Principal component analysis (95% and 99% confidence interval) and hierarchical clustering (four and six clusters) were used to find differences and similarities between tobacco products based on odor attributes.

Results:
The final attribute list contained thirteen odor descriptors. Panel consensus was reached after 14 training sessions. Attribute discrimination ability was significant for each attribute (p < 0.001). Products marketed as unflavored that formed one cluster were considered reference space. Four clusters distinguished cherry-, vanilla-, and menthol-flavored products from reference products. Six clusters additionally divided these reference products into roll-your-own, commercially available, and additive-free products.

Conclusions:
Using a trained expert panel is a successful example of assessing characterizing flavors in tobacco products. Regulatory decisions should be made on significance level and products representing the reference space, as these directly influence strictness of the method and thus the number of products that imparts a characterizing flavor compared to the reference space.