Statistical Analysis for Comparison of the Results Obtained by Capillary Columns and Packed Columns in the Determination of Water Yield in Smoke Condensates Analyzed in Cigarettes for the 24^th Asia Collaborative Study

64 laboratories participated in the 24^th Asia Collaborative Study (ACS) held in 2015. All participating laboratories are listed in Table A of Appendix A. The laboratories marked with an asterisk provided more than one dataset which were obtained by using various combinations of linear and rotary smoking machines.

In this study, 86 datasets were submitted with their water yields analysed. They consisted of 46 datasets measured by capillary column, 39 datasets measured by packed column, and 1 dataset without mention which column was used.

Participants were requested to follow the protocol “ACS” to analyse five test samples, including CORESTA Monitor CM8 (6) as listed in Table 1, and to report on parameters as listed in Tables 2 and 3.

As shown in Table 1, the four sample brands covered the range in NFDPM from 1 to 10 mg/cig, NFDPM of CM8 was reported to be 14.1 mg/cig (6). CORESTA Monitor CM8 was provided by Cerulean or Borgwaldt KC year by year in turn and Cerulean provided it in that particular year.

One dataset consists of six test results obtained from six runs, as shown in Table 2. One test result was defined as the average yield obtained from 20 cigarettes in a single run. For linear smoking machines, four ports per brand were used within each run, and five cigarettes were smoked per port. Rotary smoking machine always used 20 cigarettes per run.

Raw datasets are listed in APPENDIX B with the type of smoking machine and GC column, water yields for each run, mean and standard deviation (SD) for five test samples.

Numerical outlier technique: Cochran’s and Grubbs’ tests were applied in accordance with ISO 5725-2:1994 (7) to exclude outlying datasets prior to the determination of repeatability and reproducibility. Cochran’s test was applied to identify an outlier with statistically deviant standard deviation. Grubbs’ test was applied to identify an outlier in a univariate dataset that follows an approximately normal distribution.

After Cochran’s test, Grubbs’ test was applied to the mean values of the remaining datasets according to chapter 7.3.4.3 in ISO5725-2 (7). Grubbs’ test consists of two types of tests. One is to determine whether the largest or smallest observation is the outlier, this is called the single Grubbs’ test (chapter 7.3.4.1 in ISO 5725-2). The second test is to determine whether the two largest observations or two smallest observations are the outliers, it is called the double Grubbs’ test (chapter 7.3.4.2 in ISO 5725-2).

In the first step, the datasets were sorted in descending order for mean value. The Grubbs’ statistics Gp was calculated using the maximum mean value xp, ground mean and standard deviation s. Where the Grubbs’ statistics was higher than 1% of critical value (8), the maximum value was considered as outlier. The same process was applied for the dataset with the minimum mean value. When outliers are determined by single Grubbs’ test, the Grubbs’ test is completed. If there is no outlier in single Grubbs’ test, double Grubbs’ test for the two smallest or two largest observations should be applied.

Water yield repeatability (r) and reproducibility (R) were calculated for all types of gas chromatographic columns and each type of GC column seperately, with all types of smoking machines according to ISO 5725-2:1994 (7), by using the data that remained after the removal of outliers.

Crumpler et al. (4) also reported that there was no significant difference in water yield of CM8 between packed columns and capillary columns.

In our study, we could show that no significant difference was observed for water yields in cigarette smoke condensate between capillary and packed columns, not only from CM8 but also from other test samples, covering the majority of “tar” value products that are sold in the market.

There was no significant difference in water yields between linear type smoking machines and rotary type smoking machines for samples A and B, which had the lowest water yields. On the other hand, the water yields for samples C, D and E by rotary type smoking machines were higher than those by linear type smoking machine.

The differences in water yields between rotary and linear smoking machines were already observed in the CORESTA Harmonization Study in 1991 (10) and in the CORESTA Collaborative Study conducted by CORESTA CO Sub-Committee (11). Although the differences in CO yields and NFDPM were improved and the differences in water yields became smaller through the improvement of air flow around cigarettes, higher water yields were still observed in the results obtained with rotary smoking machines under the ISO smoking regime (4, 12). Nevertheless, such a difference is no real influence in case of evaluation of the data from each smoking machine type. Comparisons between the two types of columns were carried out with a wider range of “tar” yields for each smoking machine type. This evaluation did not show significant difference between the data from packed columns and capillary columns.

Furthermore it was confirmed that there was no significant difference in the lower range of water yields between packed columns and capillary columns, with or without distinction of smoking machine type.

The comparison of repeatability and reproducibility estimated by using all datasets as well as datasets classified by GC column type were made via an F-ratio test with a Bonferroni adjustment for multiple comparisons. No significant differences between capillary and packed columns were observed when comparing repeatability and reproducibility among the samples.

Estimated r and R listed in Table 10 were compared to those of ISO 10362-1:1999 (1) listed in Table 11 by use of 95% of upper limit of prediction bands (ULP) and lower limit of prediction bands (LLP). The ULP and LLP were determined by a linear regression analysis of mean, repeatability and reproducibility defined in ISO 10362-1:1999 (1).

The repeatability estimated by using all datasets as well as datasets classified by GC column type were plotted with those of ISO 10362-1:1999 and 95% of upper and lower limits for prediction bands (Figure 11).

Figure 11

The estimated repeatability for samples A and B were within the upper and lower limits for prediction bands. All of the estimated repeatability for samples C, D and E were smaller than 95% of lower limit for prediction band. This means that the estimated repeatability might be similar or smaller than the values in ISO 10362-1:1999.

The reproducibility estimated by using all datasets as well as datasets classified by GC column types were plotted with those of ISO 10362-1:1999 and 95% of upper and lower limits for prediction bands (Figure 12).

Figure 12

The estimated reproducibility of packed columns was almost within the 95% upper and lower limit of prediction bands. On the other hand, the estimated reproducibility of capillary columns was smaller than 95% of lower limit of prediction band. This means that the estimated reproducibility of capillary columns is better than the reproducibility in ISO10362-1: 1999.

The estimated r and R by capillary column, packed column and all types of GC column are within 95% of upper limit for prediction band.

The estimated r and R for all data, capillary column and packed column are confirmed to be comparable to those of ISO 10362-1:1999, and these values can cover a wider range of water yields by ISO Smoking Regime.

The comparative results of water yields and estimated r and R strongly suggests that capillary columns are an appropriate alternative column for the gas chromatographic procedure of ISO 10362-1.