HPHC Testing of Tobacco and Smoke to Examine Cigarette Temporal Variability

The study was conducted in three phases to evaluate short-term, medium-term, and long-term product variability (see Table 1). Results were not compared across products; the design was effectively eight parallel independent studies. For Phase 1 of the study, Batches A, B, and C were collected within 1 week of production. A larger sample size for Batch B was collected to include this batch in each phase of the study (B1, B2, B3). For Phase 2, Batches D, E, and F were collected approximately quarterly with Batch B, labeled B2, treated as the first quarter batch. For Phase 3, Batches G and H were collected annually with Batch B, labeled B3, treated as the first-year batch. Quantities of a single batch of 3R4F were collected, labelled, stored, and shipped similarly to the product batches. The choice of collection timing was based on trying to gradually incorporate expected and unknown production variables into the study. It was anticipated that ‘one-week’ would allow for varied shift, equipment, operator, and raw material changes such as cigarette paper bobbins or filter tow bales. Quarterly collections for one year (Phase 2), was expected to include these typical production variables and possibly grade changes or blend turn-over depending on a given manufacturer's undisclosed production practices. Collecting product across three years of production, was expected to magnify each of these and other likely production variables.

The scope of the study was limited to commercial factory-made cigarettes in order to create ‘real-world’ information. Reference cigarettes (3R4F) were incorporated for comparison purposes. Considerations included: region of the world (United States, Europe, China); blend (American, Virginia, Dark Air Cured); non-intense “tar”-level (<1–20 mg/cigarette); and filter construction. Mentholated products were excluded due to complexity with cigarette conditioning. The first certified reference product for cigarettes, 1R6F, came available during the course of the study, and was added to the study for informational purposes.

While no attempt was made to control or influence raw material lots or manufacturing schedules, product volunteers agreed to provide a relatively high-production-volume product to avoid limited production runs and low raw material turn-over. Specific manufacturing information was not collected due to confidentiality considerations between different manufacturers.

Two Virginia and six American blend commercial products were chosen to represent a range of design parameters (e.g., blend type and “tar” level, see Table 2). Virginia blends are typically all flue-cured tobaccos, often with no or limited added flavors. American blends typically contain flue-cured, Burley, and Oriental tobaccos with or without added flavors.

Product collections were from typical commercial production runs with no additional controls imposed. Samples were collected according to the time schedule (Table 1) and stored at −20 °C prior to shipment to the testing labs. No significant design changes were noted for the study products during the course of the study.

One batch of 3R4F cigarettes was included as a study control to compare testing of multiple batches of production samples with repeat testing of a single batch of 3R4F cigarettes treated to the same study design parameters.

For data collation, the test samples were distinguished by Sample-Batch designations such as 1-B1 (corresponding to Batch B1 from product Sample 1).

HPHCs tested were based on recommendations or requirements of regulatory bodies (World Health Organization, U.S. Food and Drug Administration, Health Canada, Chinese State Tobacco Monopoly Administration).

Additional measures as indicators of analytical quality, such as total particulate matter (TPM), puff count, and conditioned cigarette weight, were reported as well. Physical measurements were included as indicators of cigarette production consistency.

This set of measures afforded the potential for trends analysis based on attributes such as matrix (leaf and smoke); constituent phase (particulate, semi-volatile, volatile); and relative level of constituent yield (ng, μg, mg).

Testing volunteers were solicited from among industry laboratories with capabilities for TNCO and at least one constituent class in the test list. Laboratories were typically ISO 17025 accredited and it was recommended but not required for the laboratories to use published or standardized analytical methods.

All sample batches were shipped and tested concurrently within each phase of the study. For each constituent and each phase, a single lab tested all of the samples (n = 5). Five replicates provided approximately 80% power to detect a ratio of the sample-to-sample standard deviation to the replicate-to-replicate standard deviation of 1.32 or more which was deemed adequate power (28). Due to the large number of replicates for a given phase, testing was expected to take from several days to several weeks to complete. Therefore, replicates were interleaved to minimize the confounding of laboratory drift over the testing time with the manufacturing variation under study. Due to interleaving, repeat analyses were not required except in cases of fewer than 3 reportable results. Due to shipping complexities across multiple countries, not all product samples were received by all testing laboratories in Phase 3. In these cases, results are noted in the data tables as NCS (not calculated due to shipping).

The TNCO testing design varied from the other analytes of measure. Rather than one laboratory, all laboratories tested TNCO concurrent to testing of their assigned analyte class.

Certain study design features were put in place to minimize analytical variability. Batches were collected, held at −20 °C, and shipped together for testing to avoid temporal variability during testing. Except for TNCO, all samples were tested for a given method in a single laboratory at the same time to avoid lab-to-lab variability or within lab temporal variability. Replicates were interleaved to prevent bias in analysis due to analytical drift.

The laboratories reported key information such as TPM, puff count, and conditioned cigarette weight to assess the data for obvious anomalies that may be harder to discern in analyte results. TNCO was carried out in all the laboratories and provided additional insight into the testing variability of the products.

Unidentified or unknown contributors to analytical variability were naturally accounted for in the analysis through inclusion of repeat testing of a single production batch of 3R4F as a study control and through the repeat testing of the “B” batches in each study phase.

Moisture, cigarette weight, tobacco weight, filter tip ventilation, circumference, length, paper permeability, and resistance to draw were measured using conventional equipment in common use in industry laboratories (29, 30, 31).

Tobacco was removed from the other cigarette components in preparation for analysis. As warranted, tobacco samples were oven-dried for consistent grinding; thus, analytes may be reported ‘as-is’ or on a dry-weight basis. Replicate analysis was conducted on separate-grind aliquots. -

Nicotine: Ground tobacco was extracted using liquid/liquid extraction into organic solvent containing n-heptadecane as an internal standard followed by GC-FID analysis according to the “hexane method” for CRM 62 (32).

Mainstream smoke was generated under both ISO 3308 (non-intense) (10) and ISO 20778 (intense) (11) machine smoking regimes. Cigarettes were conditioned before smoking in accordance with ISO 3402 (36).

-

“Tar”, nicotine, carbon monoxide (TNCO): Cigarettes were smoked onto glass filter pads which were subsequently extracted with alcohol, extracts were analyzed by GC-FID and GC-TCD. Carbon monoxide was analyzed by in-line gas trapping and NDIR (10, 11, 37, 38, 39, 40).

Data were compared using one-way analysis of variance, except TNCO which was two-way including interactive effects. The different handling of TNCO was because multiple labs analyzed the samples for TNCO and only a single lab analyzed the samples for other HPHCs. Obvious outliers noted in the raw data supplement were excluded by visual inspection.

The range of the observed batch means for a set of timepoint batches was used to aid in the interpretation of whether the differences in batch values were of practical importance. Range=(maximum batch mean− minimum batch mean)/average of means of all batches {\rm{Range}} = \left({{\rm{maximum}}\,{\rm{batch}}\,{\rm{mean}} - \,{\rm{minimum}}\,{\rm{batch}}\,{\rm{mean}}} \right)/{\rm{average}}\,{\rm{of}}\,{\rm{means}}\,{\rm{of}}\,{\rm{all}}\,{\rm{batches}}

The percent differences among the batch timepoints across the entire study, as displayed in Supplemental Table 1, ranged from 0.1% to 27.1% but were typically below 10%. Results for 3R4F ranged from 0.1% to 14.0% and were typically below 5%. Filter tip ventilation, resistance to draw (open/closed), and paper air permeability are shown to have the highest percent difference among the batch timepoints for the measurements. Several of the products had statistically significant differences for the timepoints.

Nicotine, ammonia, TSNAs, and trace metals were determined for tobacco samples. TNSAs were determined by the same testing laboratory that measured smoke TSNAs for matched analytical methodology.

Arsenic and cadmium levels were near or below the lower limits of quantitation for many of the samples which resulted in non-reportable values or artificially elevated percent differences. Thus, these analytes are not useful for the study objective and the variability calculations are not reported but rather are noted as not calculated (NC).

For NNN and NNK, many of the products appear to have batch-to-batch variability above the level shown by 3R4F (repeat testing of a single batch) with apparent trending based on study phase. NNN product variability averaged approximately 10%, 40%, and 28% for the products’ short-term, medium-term, and long-term comparisons, respectively.

Repeat testing variability for 3R4F for the same study phases was 3.8%, 2.8%, and 3.2%. Results for NNK followed a similar trend: products’ variability averaged approximately 14%, 41%, and 29% for the three study phases. 3R4F showed slightly greater repeat testing variability than for NNN at 3.3%, 9.2%, and 8.9%, respectively.

Tobacco nicotine differences were slightly greater for product batches (~ 3%) collected within one week compared to repeat testing of 3R4F (0.4%). Differences were slightly greater and were statistically significant for 1-year and 3-year studies at approximately 6.8% and 5.7% on average for products compared to 1.7% and 0.1% for 3R4F. The highest difference for tobacco nicotine noted among the timepoints was 14.0% for the Product 3 long-term difference.

Like TSNAs, the tobacco ammonia results were often quite different from one another in the medium and long-term batches. The Virginia blend cigarettes (Products 3 and 8) sometimes showed quite large percent differences (> 60%) because their levels were low; small changes could result in large percent differences. For the American blend cigarettes, short-term ammonia differences among batches were greater for most products than for 3R4F (3%). Most of the American blend products showed somewhat greater differences than 3R4F in both the medium and long-term (13% on average compared to < 5% for 3R4F).

Mainstream smoke was collected using non-intense and intense smoking regimes for the determination of ammonia, BaP, carbonyls, PAAs, TSNAs, HCN, and VOCs. For each of these analyte classes, all testing was conducted in one laboratory and all samples for a given study phase were tested concurrently.

On average, across all products and all analytes, the product differences for non-intense smoking (15%) were greater than 3R4F repeat testing differences (7.2%). Intense results are similar; on average, sample differences over time are slightly greater than repeat testing of 3R4F. The differences between samples under the intense smoking regime were low except for TSNAs. This supports previous conclusions regarding ventilation's impact on variability for non-intense smoking (26). Results of particular interest are noted below.

Product 7 with high filter tip ventilation showed relatively larger differences for the non-intense regime (as high as 106.4%). This is a very low yielding product; several analytes were at or below method limits of quantitation (LOQ) and small differences resulted in large proportional differences.

Ammonia results for Product 3 and 8 (Virginia blend cigarettes) appear to show trending for study phases. For Product 8 intense regime, Phase 1 difference is 13.4% compared to 7.2% for 3R4F and Phase 2 is 45.2% compared to 14.1%. As with trace level tobacco metals, variability among the replicates is relatively high with some values at or below the method LOQ for non-intense regime results. Phase 3 testing for Product 8 was not completed due to product distribution issues. Product 6 ammonia differences also appear to trend up for medium- and long-term sampling compared to collecting all batches within a week. Ammonia is a naturally occurring constituent yet may also be an additive (e.g., processing aid) in tobacco and/or cigarette paper.

This is an unknown for the study samples since these types of manufacturing details were not disclosed. Thus, smoke ammonia variability may be influenced by natural and/or added content.

As with tobacco analysis, NNN and NNK yields for non-intense smoking showed statistically significant differences of modest size in many products for the short-term study (~ 15% compared to ~ 4% for repeated 3R4F) and relatively larger magnitudes of difference for the later phases (as much as 62.9% compared to ~ 13% for 3R4F). These compounds are known to be related to the tobacco levels of those HPHCs, and they are also known to vary widely in tobacco due to agricultural variation over time. This trend held for intense smoking as well.

Unlike other smoke analytes, TNCO testing was conducted in all laboratories.

Batch-to-batch variability for each of the samples was low for each of the phases and all samples (~ 5%). Conversely, lab-to-lab variation was statistically significant in all but two comparisons across all TNCO samples, analytes, and regimes.

There are also potential interactive effects (e.g., Sample × lab) noted in the data. Table 4 shows the example of Product 2, non-intense CO yields for the Phase 3 batches. All batch-to-batch differences compared to lab-to-lab differences of these HPHCs are provided in the supplemental materials.

There were no obvious variability trends in the data due to blend type (American or Virginia) except that Virgina blends tend to be low in ammonia (51) causing low or < LOQ values. High filter-tip ventilation primarily influenced analytical variability and the carbon filter additive affected VOC yields, but neither of these product design variables was observed as a major factor influencing batch-to-batch product variability.

HPHCs tested were based on recommendations or requirements of regulatory bodies (World Health Organization, U.S. Food and Drug Administration, Health Canada, Chinese State Tobacco Monopoly Administration). Additional measures as indicators of analytical quality, such as total particulate matter (TPM), puff count and conditioned cigarette weight, were reported as well. Physical measurements were included as indicators of cigarette production consistency.

This set of measures afforded the potential for trends analysis based on attributes such as matrix (leaf and smoke); constituent phase (particulate, semi-volatile, volatile); and relative level of constituent yield (ng, μg, mg).

As previously observed by Agnew-Heard et al., cigarettes with relatively high variability of physical parameters tended to show relatively greater variability for non-intense smoke yields (52). Constituents that transferred from leaf to smoke showed similar trends for batch-to-batch variability trending. Both leaf TSNAs and smoke TSNAs showed a greater batch-to-batch variability than repeat testing for 3R4F and showed greater differences among batches for the longer-term studies. This trend was also observed for ammonia and, to a lesser extent, nicotine. There were no obvious trends noted based on constituent phase or relative yield.

Our objective was to expand our understanding of commercial cigarette product variability in a more holistic manner. The use of statistical comparisons was employed to aid objectivity in our data analysis and in highlighting more complex trends such as the interactive effects noted for TNCO results. A review of Table 3 and Supplemental Table S1 will reveal that some batch differences that were statistically significantly different were extremely small and of little practical concern. For that reason, we included percent differences in addition to statistical significance to be able to judge whether the observed differences were meaningful from a practical perspective.

As an additional safeguard, we included the 3R4F reference product in the study. The testing design for the 3R4F was the same as for the commercial products so that we could ask the question for each set of analyses: “Is the difference observed for testing multiple batches of Product X greater than that observed from repeat testing of one batch of the reference?”. If not, even if relatively high, the difference observed is likely not due to product (batch) differences but more likely due to analytical variability.

Like 3R4F, B Batches were also single production batches that were repeat-tested. In this case, they were each made in the first week of production but were stored, shipped, and tested with each phase of the study, and hence are also a type of design-matched control for associated samples. Though not evaluated in this data analysis, one may be able to estimate temporal analytical variability for the study using the Batch B data and/or the 3R4F data from all of the phases.

Basic physical parameters were included in the study to understand manufacturing consistency for the study samples. If cigarette weight were highly variable, for example, high smoking yield variability could well follow. Most of the parameters tested were similar in variability for the products to the variability observed for 3R4F repeat testing. Some examples of greater variability for filter tip ventilation were noted. In particular, Sample 7 with a carbon-containing filter of high ventilation (~ 80%), showed a percent difference in batches of 14–20% across the three phases compared to approximately 5% for each of the phases for 3R4F.

None of the measures appeared to show differences in magnitude based on study phases; short-term, medium-term and long-term variability were similar and low for all the parameters. This is an indicator of a high degree of manufacturing process control and likely relatively low product manufacturing variability. Cigarette weight, probably the most fundamental indicator of production variability, ranged from 0–4% difference between batches of products collected across 3 years of production.

One of the main reasons to divide the study into phases of < 1 week to as long as 3 years was an attempt to incorporate tobacco crop-year changes. Metals data did not support a trend analysis as they typically were near or below the method's lower LOQ. This highlights a real-world issue with reporting requirements for trace or < LOQ analytes. NNN and NNK yields, conversely, showed clear temporal batch-to-batch variability. Samples collected within 1 week, in some instances, had significant variation (27%), and even greater variation over longer periods (82% collected across 1 year's production). Repeat testing of 3R4F (3%, 4%) supports that the differences are product-related rather than analytically driven. Repeat testing of the product's Batch B further supports this and confirms consistency within a batch as illustrated in Figure 1 showing tobacco NNN for Product 9 and 3R4F across all batches. Interestingly, the data seem to indicate that medium-term variability is higher for NNN and NNK (~ 40% on average) than for long-term variability (~ 30% on average). It is possible that there is an in-common but unknown raw material or manufacturing variable that is influencing these results. Or, perhaps ~ 40% and ~ 30% are generally equivalent. Repeat testing of Sample B and of 3R4F confirmed that this is not an analytical artifact. A study conducted with additional timepoints may be warranted to better understand this trend.

Figure 1

Evaluation of puff count or TPM were included as indicators of consistency of smoking, regardless of constituent. An average of all puff counts across the study showed < 5% difference.

For constituent evaluation, data for combustion-generated HPHCs showed low relative percent differences for batch comparisons. See Figure 2 for Product 1 non-intense BaP results as an example. Sample variation for the study phases were of a similar magnitude as replicate variability and approximately 5% or less compared to < 2% for repeat testing of 3R4F with each phase. The results support that combustion byproducts are primarily influenced by amount of tobacco burned and cigarette design variables such as level and variability of filter tip ventilation.

Figure 2

Conversely, tobacco-related compounds, such as NNN and NNK, tended to show larger differences in yields between batches collected over a longer period of time (1 year or 3 years) as compared to samples collected within 1 week. It follows that tobacco HPHCs vary naturally over time and thus vary in smoke over time. Tobacco HPHCs, whether tested in smoke or tobacco, show large temporal variability.

Smoke ammonia may result from a mix of sourcing from combustion and from transfer from tobacco (22), whereby products’ smoke ammonia yields had larger batch-to-batch differences for batches over longer periods but did not necessarily correlate as strongly between smoke and tobacco results as TSNAs depending on the product.

TNCO was originally intended to provide more information about the samples and not to serve as a laboratory comparison. Our rationale was that all of the participating laboratories have full capability and years of experience with these standardized methodologies leading one to expect lab-to-lab variability to be low and therefore allow for additional data points for TNCO. However, the TNCO lab-to-lab variability was found to be much higher than batch-to-batch product variability and was shown to not trend with study phases.

A few potential implications show themselves in the take-aways from the TNCO analysis. First, this part of the study was conducted more like a typical interlaboratory study and shows a typical level of variability among the datasets. Second, different laboratories testing the same samples in the same timeframe showed a range in nicotine from, for example, 0.58 to 0.84 mg/cigarette for the same sample. This large of a spread in reporting is a caution against over interpretation of single-point results. Also, differences could be even larger for more variable or less well-established analyte classes. Additionally, more formal evaluation of TNCO on this scale would likely warrant sampling in compliance with ISO 8423 (53), which was considered to be overburdensome for this study due to the study size, and data analysis for tolerances in compliance with ISO 22305 (54).

In this study, we purposely stored production samples to test them concurrently to minimize analytical variability. We also isolated testing to single laboratories except for TNCO. We found that the variability in the data was relatively low compared to studies for which products are controlled in order to study analytical variability. The one aspect of the study, TNCO testing, that was designed in a more typical manner, had more typical (i.e., higher) variability among the data sets. Thus, the primary implication of the results of the study is that analytical variability is likely the major contributor to the overall reporting variability in a typical study.

On the other hand, to facilitate on-going product testing for whatever the reason, production, product, and analytical variability often cannot be isolated in this way. Thus, in real-world testing product variability and analytical variability will remain confounded.

In evaluation of data sets from different laboratories, one must consider that all are subject to these same analytical variables and may use different analytical procedures.

While one is not able to eliminate analytical variability, reporting of the same control/reference samples across studies and between laboratories can aid evaluation of the data.

The number of samples of each product is limited and are not sufficient to provide good quantitative estimates of variability. For example, in order to ensure that the testing volumes were manageable, the formal sampling procedures outlined in ISO 8243 were not followed. Thus, the estimates of product differences should be regarded as illustrative rather than quantitative.

Due to the sheer size of the study, additional relevant study variables such as lab-to-lab variability for a given method and the impact of sample storage conditions were not included and cannot be assessed from the data. Additionally, a nested design would be more ideal from a statistical analysis perspective but too large to have been achievable.

The study design also does not allow for product-to-product comparisons. Effectively, the design created 8 independent studies conducted in parallel. Our initial discussions of the study design included production of a set of cigarettes designed specifically for the study to be produced by each manufacturer; to hold designs/recipes as constants. We decided against this, as a study of this design would just confirm how well a manufacturer could make cigarettes for which they have no experience and un-optimized equipment. That design would not have measured actual commercial cigarette variability.

Samples were not controlled for raw material and/or cigarette making equipment used. Product information such as blend composition or inventory practices were not disclosed. Thus, specific factors related to product variability, other than time of production, cannot be isolated.

Data analysis was limited to the direct scope of the study objectives. There may be other viable analysis questions and analysis tools applicable to the data set.