Reference change values to assess changes in concentrations of biomarkers of exposure in individuals participating in a cigarette-switching study

Clinical study

The clinical study (ISRCTN72157335) was conducted in Hamburg, Germany, in 301 healthy adults smokers (minimum age 21 years) who had smoked for >3 years, had smoked their current brand for >6 months and typically smoked between 10 and 30 cigarettes per day. The study was designed and conducted in accordance with the ethical principles of the Declaration of Helsinki and the International Committee on Harmonization for Good Clinical Practice. Details of the study design and protocol have been reported elsewhere [14]. Briefly, volunteer smokers were supplied with unbranded commercially available cigarettes similar to leading commercially available brands in Germany in 2007–2008, with the same International Organization for Standardization (ISO) tar yield as their usual brand (1 mg or 6 mg) for 2 weeks. On Day 14, participants entered the clinic for a short period of clinical confinement, during which cigarette consumption was recorded and 24 h urine samples were collected to measure baseline urinary levels of BoE to tobacco smoke toxicants. Two groups of subjects continued to smoke the 1 mg and 6 mg commercial cigarettes for the remaining 4 weeks of the study (commercial control groups CC1 and CC6, respectively), whereas the others were switched to an RTP cigarette with the same ISO tar yield as their usual brand (one 6 mg ISO tar yield RTP, TSS6, or one of two 1 mg ISO tar yield RTPs: TSS1, which contained tobacco-substitute sheet in the blend and high activity carbon and amine-functionalised resin in the filter, or BT1, which contained water and enzyme-treated Virginia tobacco in the tobacco blend and the same filter as TSS1). These products are described in greater detail elsewhere [15, 19–22]. Urine samples were collected and cigarette consumption recorded in two further periods of clinical confinement (Supplemental Data, Figure 1, which accompanies the article at http://www.degruyter.com/view/j/cclm.2014.52.issue-3/issue-files/cclm.2014.52.issue-3.xml). The final samples collected on Day 41 were used to assess the endpoints of the study. Non-smokers (not smoked for >5 years and baseline urinary cotinine levels <10 ng/mL) were enrolled to provide background levels of urinary BoEs from environmental and dietary sources. The ratio of male-to-female subjects was between 3:2 and 2:3, and groups were matched for age.

Twenty-one known BoEs for tobacco smoke toxicants with validated analytical methods [13, 23–28] were assessed in total (Supplemental Data, Table 1). Levels were reported as quantities of BoEs excreted over 24 h, and were calculated as products of the concentrations in urine and the volumes of urine produced by the subjects over the collection period. Statistical analyses have been carried out using amount of BoEs adjusted by cigarette consumption. Reasons for using this approach are stated in the “Statistical analysis endpoints” section. Four of the investigated biomarkers, NNAL, N-nitrosoanatabine (NAT), 3-hydroxypropylmercapturic acid (HPMA) and 3-hydroxy-1-methylpropylmercapturic acid (HMPMA), are used to illustrate the methods of this report.

Statistical analysis endpoints

To investigate the number of subjects driving observed changes in the original group analyses, we initially needed to define what constitutes a significant change. For simplicity and consistency with previous analyses, hereafter change is defined as BoE level on Day 41 (end of study) minus the level on Day 14 (except for in the non-parametric method), and can be calculated as

Endpoint_i=BoE day 41_i–BoE day 14_i

where i represents a subject. Figure 2 of the Supplemental Data illustrates the distribution of these differences across smoker groups for HPMA. Statistical analyses were conducted with SAS version 9.3, and graphics were developed with JMP 10 (both from SAS Institute, Cary, NC, USA), unless stated otherwise.

The endpoint of change can be expressed as two different magnitudes: first as the total amount of a BoE (excreted in 24 h), and second as the total amount of a BoE adjusted by the number of cigarettes smoked per subject. As the two magnitudes yield similar results (Figure 1), we have focused on the adjusted magnitude for this study, as this approach accounts for differences in levels of cigarette consumption. The numbers of cigarettes smoked by subjects were estimated as the mean values of the self-reported daily consumption and the number of tips collected during confinement periods of the clinical study. As temporary variations can lead to disparities between self-reported and counted consumption, the mean values of those quantities is believed to give the best estimate of cigarette consumption per subject.

Figure 1

Amount of NAT excreted on Days 14 and 41 (top) and amount adjusted by the number of cigarettes smoked (bottom).

NAT, N-nitrosoanatabine.

Reference values

Increases or reductions in BoE levels within individuals might be the result of natural variability rather than a direct consequence of switching to an RTP. This variability will be the sum of analytical variability of methods used plus biological variability within individuals. We used several methods to calculate reference values to assess suitability of statistical approaches to monitor changes at the individual level.

Empirical rule

We compared results from the non-parametric approach and its interpretation with those obtained by the empirical rule of statistics. In this context, to apply the empirical rule we departed from the hypothesis that no change would give a mean of zero, and we estimated the standard deviation (SD) for each group (SD_j) from the changes within groups to calculate reference limits by group (j=TSS6, CC6, BT1, CC1 and TSS1). Therefore, the mean is set to 0 and the reference limits are set at 2 SD_j and 3 SD_j from that reference. Subjects yielding responses >3 SD_j or <−3 SD_j are considered to have experienced very significant changes. Changes between 2 SD_j and 3 SD_j or −2 SD_j and −3 SD_j are deemed significant, and other responses (between 2 SD_j and −2 SD_j) are classified as no change. Figure 2 illustrates how differences between time-points are represented and their empirical reference limits by group with their means set at 0.

Figure 2

Differences between Days 14 and 41 with empirical reference intervals by group of smokers.

The mean is set as zero and the reference lines at set at 2σ and 3σ. HPMA, 3-hydroxypropylmercapturic acid.

Reference change values

Although empirical reference limits provide information at the individual level within groups, the generalisability of results is limited. An improvement to this approach is the use of RCVs. The RCV method is conceptually similar to the empirical method. Individual subjects’ BoE levels are deemed to be stationary at a specific level and some natural or random variation will occur over time. RCVs were first introduced by Harris and Yasaka [31] and later developed by Fraser and Harris [32, 33]. This methodology provides reference values based on within-subject (CV_I) and between-subject (CV_G) biological variation and, therefore, enables dissociation between analytical and biological variability. RCVs are widely implemented to help clinicians in the interpretation of laboratory tests intended to monitor subjects over time and/or to assess the prognosis of diseases [34, 35]. This methodology enables incorporation of more than two time-points into the analysis to calculate the sources of variability. Calculations of reference values are based on data collected on Days 14, 28 and 41 in the two commercial cigarette control groups.

The overall variability is assumed to have two main sources: within- and between-subject variability. Within-subject variability can be seen as random variation of values from homeostatic biomarker levels in response to environmental factors per subject and as variations in analytical methods over time. For this report, variability within and between subjects was calculated with repeated-measures analysis, by use of PROC MIXED software, with subject as a random factor. Following the recommendations of Fraser and Harris [32], possible outliers were examined. Few observations for each biomarker were identified as statistical outliers, but, given that outlying values did not consistently belong to any specific subject across biomarkers and there was no evidence of clerical error, we decided not to remove any observations from the analysis, as recommended by the CLSI [30]. Homogeneity of variances was accepted and a Gaussian (at least symmetric) distribution assumed. Homogeneity of variances was also examined with the index of heterogeneity [32].

Within-subject variability (CV_I) was estimated as the within-subject mean sum of squares. The between-subject variability (CV_G) was calculated as the between-subject mean sum of squares minus the within-subject mean sum of squares, divided by the number of measurements per subject. The RCVs were calculated two-sided as 1.96×√2(CV_I²+CV_BA²) × overall mean of endpoint, where CV_BA is the between analytical batches variation [36, 37]. Categorisation of subjects as experiencing change or not were performed with respect to the RCVs in a 2-week period, i.e., Day 28–Day 14 and Day 41–Day 28.

Sub-group analyses

We assessed whether changes in BoEs could be detected in groups of subjects not considered previously. The explanatory variables assessed in the clinical study as possible determinant factors that could have an impact on RTP effects were sex, age and body mass index. Relationships between the statistical endpoints in this study and these explanatory variables were assessed with general linear-model analyses. Variables and the smoker groups were modelled in pairs. Three general linear models, each of which included as variables smoking group, one of the explanatory variables and interaction, were used to find possible relationships by BoE, for which we calculated F-statistics for each variables and their interactions terms. Models containing only main effects were assessed if an interaction effect was found not to be statistically significant. Larger models were deemed inappropriate due to small sample size. The study, however, was not designed to include this type of analysis, which limits generalisability of results and leads to lack of statistical power. Thus, the analyses should be viewed as data explorations that might lead to future studies. Model assumptions were visually assessed and were considered acceptable given the limitations previously stated. Tukey’s HSD post-hoc test was used to assess differences between the levels of the significant parameters in the regression model.

Non-parametric references

In an attempt to identify subjects who experienced changes in BoE levels greater than the expected biological variability, we calculated non-parametric 95% CI at baseline and classified subjects with respect to these references at Day 41.

Hence, non-parametric 95% CI were calculated for the 95 and 133 subjects who at Day 14 were smoking, respectively, 6 mg and 1 mg ISO tar yield control commercial products (Table 2). The CIs are systematically higher for the 6 mg tar band than for the 1 mg band, with the exception of those for NAT, for which the intervals are very similar.

For each 95% boundary (reference), 90% CLs were calculated as a measure of uncertainty of these estimates. By visually comparing estimates (Table 2), differences in BoE between tar band references were suggested in the upper boundary (labelled HIGH). HMPMA upper bound CLs do not overlap between tar bands, which suggests that smokers of cigarettes in a higher tar band reach significantly higher yields than those who smoker lower yield cigarettes. The limits for the lower bound references (labelled LOW) do not suggest differences between tar bands. For HPMA and NNAL the upper boundary CLs overlapped slightly between tar bands but references are outside the 90% CL of the other tar band, while no differences between tar band references for NAT were suggested.

We used the non-parametric 95% CI to classify changes at Day 41 as increases if subjects yielded BoE levels above those limits, decreases if BoE levels were below the interval lower boundary, and not changed if BoE levels remained within the intervals estimates. Frequencies show that levels did not change for most of the subjects (Table 2). For NNAL and NAT, at least 89% of subjects were within the intervals, which is in contrast to the decreases of up to 100% seen for BT1 when only the direction of change of the endpoint was assessed (Table 1). Patterns of decreases for HMPMA are still clearly visible with 65%, 49% and 64% of subjects being classified as having decreased BoE levels with TSS6, BT1 and TSS1, respectively. Some increases for control groups were also suggested with increases of 10% or more for subjects in the CC1 group with respect to three of the biomarkers.

Frequencies generated by the non-parametric approach are difficult to interpret, given that 95% CIs are calculated from a snap-shot at baseline and, therefore, do not take into account changes at the individual level. If the level of a BoE at Day 14 is close to or outside a boundary, no change or a small change relative to other individuals could appear as an increase or decrease, whereas substantial changes in individuals with Day 41 values within the boundaries would be categorised as no change.

Following this reasoning, we could have calculated the 95% non-parametric CIs of the differences (Day 41–Day 14) or within-subject variability. On this occasion, however, we can only use data for subjects in the control groups (n<50). Thus, we were precluded from using this approach as the CLSI recommends use of at least 120 observations.

Empirical rule

As an alternative to non-parametric CIs, the empirical rule was used to calculate approximations of the references based on the differences of BoE levels between Day 14 and Day 41, we assumed that these differences are symmetrically distributed around the mean.

With the empirical reference intervals centred at 0, differences in BoE levels between Days 14 and 41 were assessed with respect to 2*SD and 3*SD by smoking group. Although general patterns are maintained, the number of subjects classified as experiencing change is significantly different to numbers obtained with previous approaches (Supplemental Data, Table 2).

Evaluation of the direction of BoE changes showed that 94% of subjects experienced reductions in NNAL levels in group that received the BT1 RTP (Table 1). With use of the empirical rule references, however, 75% of smokers fall within the normal expected variability (no change) and the remaining 25% experienced significant reductions in NNAL levels (Supplemental Data, Table 2). Increases are also suggested for some control groups, with 13% of subjects experiencing significant increases and 4% very significant increases for HPMA in the CC6 group. HPMA and HMPMA were both shown by group analysis to have reduced levels in the BT1 RTP group, but the empirical rule criterion showed that 15% of subjects had significant decreases and 4% had very significant decreases for HPMA, whereas for HMPMA the decreases were significant in 38% and very significant in 7% (Supplemental Data, Table 2).

Reference change values

RCVs are used to assess change for two consecutive measurements, but they can be based on more than two time-points by calculation of within- and between-subject variability. The methodology also enables dissociation of within-subject variability from analytical variability.

Biomarker data collected from subjects in the commercial control groups at Days 14, 28 and 41 were used to determine subject variability for the two tar bands (Table 3).

Differences between two serial measurements larger in absolute value than the corresponding RCV are assessed as change. On the basis of these RCVs, change frequencies were calculated by tar band and all smoking groups (Supplemental Data, Tables 3 and 4).

Although results generally seem to be more conservative with the empirical rule method, overall similar patterns emerged from that and the RCV approach. Most changes seemed to occur within the first 2-week period (Days 14–28), which indicates a rapid response of the levels of these biomarkers to changes in cigarette smoke. If it is accepted that BoE levels reached a homeostatic set-point within the study period, the frequency of changes should be comparable with the total frequencies obtained when the endpoint BoE level on Day 41 minus that on Day 14 is used (Table 4).

Discrepancies between numbers of observations for the BT1 RTP in different tables are the result of missing data for one subject on Day 28.