Health Canada
www.hc-sc.gc.ca
Common menu bar links
Institutional links
Priority Substances List Assessment Report- 1,3-Butadiene
3.0 Assessment of "Toxic" Under CEPA 1999 (Continued)
3.3 CEPA 1999 64(c): Human health
3.3.3 Exposure-response analyses
Since air is the principal route of exposure to butadiene in the general environment (available data indicate that other routes contribute negligibly), quantitation of exposure-response for cancer and non-cancer effects is limited to exposure by inhalation.
In order to eliminate the uncertainty associated with extrapolation from animal species, quantitative measures of carcinogenic potency (i.e., tumorigenic concentrations, or TCs)4 have been developed on the basis of available epidemiological data. This is based on the conclusion that the weight of evidence for an association between butadiene and leukemia satisfies several of the traditional criteria for causality in epidemiological studies. However, uncertainties in the exposure estimates for the critical cohort of workers as well as confounding or effect-modifying aspects that could impact on quantitative estimates of risk are recognized. In view of these factors and to serve as a basis for comparison, quantitative measures of cancer potency have also been developed on the basis of results of long-term bioassays in rats and mice, with those in mice being considered justifiably conservative, considering the likely heterogeneity in metabolic transformation of butadiene in humans. (See discussion of relevance of specific tumour types in animals to humans in Section 3.3.3.1.2.)
In addition to inducing tumours at multiple sites in experimental animals, there is also sufficient evidence that butadiene is genotoxic in somatic and germ cells and induces reproductive and hematological effects in animals. As a measure of exposure-response for noncancer effects, where considered appropriate, benchmark concentrations5 have been calculated on the basis of data from long-term studies in mice.
Several physiologically based pharmacokinetic (PBPK) models have been developed as a basis for reducing uncertainty in interspecies extrapolations for butadiene by various groups of investigators. However, none of the models currently available has adequately accounted for the distribution of metabolites in the compartments included; the principal researchers in this field have concluded that there are likely more factors involved in butadiene metabolism than have been included in the models developed to date (Csanády et al., 1996; Sweeney et al., 1997). In addition, none of the models has included the formation of EBdiol, a putatively active metabolite that is believed to be important in humans, since it has been observed to bind to hemoglobin to a greater degree than EB in workers exposed to butadiene. Nor has bone marrow been incorporated as a compartment, although it appears to be a target site of butadiene-induced toxic effects. Moreover, none of the PBPK models has been validated in humans. For these reasons, therefore, such models have not been used to quantitatively account for interspecies variations in metabolism in the quantitation of exposure-response for critical endpoints based on studies in experimental animals presented here. In addition, owing to its relatively slow metabolism, butadiene achieves a steady state during prolonged inhalation exposure. On this basis, exposures of the same concentration and duration would be expected to result in equivalent toxicity across species, and no interspecies scaling to account for variations in inhalation rate to body weight ratios or body surface areas between humans and animals have been incorporated.
3.3.3.1 Carcinogenicity
3.3.3.1.1 Estimated potency based on epidemiological data
In only one epidemiological investigation of the association between butadiene and leukemia have data on exposure of the study population been sufficiently characterized to permit quantitation of exposure-response (Delzell et al., 1995). The Delzell et al. (1995) study also presents results for the largest cohort studied to date (including subjects from eight plants, six of which were included in the exposure-response analyses); it is also considered to subsume the observations of mortality in workers at these plants reported previously by other researchers (i.e., Meinhardt et al., 1982; Matanoski et al., 1990, 1993; Santos- Burgoa et al., 1992), because of the considerable overlap in the cohort definition. The exposure assessment of study subjects was of extremely high quality, being very thorough and based on industrial hygiene monitoring data (although limited and used primarily for comparison with estimated concentrations), research of plant records concerning work histories, processes and local emissions, and consultation with staff from each plant, and is, therefore, considered appropriate for quantification of exposure-response. For comparison with estimates based on the data from the cohort study, carcinogenic potency was also calculated on the basis of the results of the case-control study nested within essentially the same population of workers (Matanoski et al., 1997), although data available in the published report were too limited to permit detailed analysis here.
Methods
The raw study data6 for the six plants investigated by Delzell et al. (1995) were used to calculate the potency estimates. The data consisted of the cumulative occupational exposures to butadiene and styrene at each year of each subject's life (person-year), beginning with his entry into the cohort and terminating with death or other exit from the cohort. The data also contained information on race, age, calendar year and years since hire of each subject.
The response of interest was cases of death due to all forms of leukemia, as information on the specific type of leukemia was insufficient; only cases in which leukemia was considered the underlying cause of death were considered in these analyses. Exposure estimates were cumulative occupational exposures in ppm-years assumed to be incurred for 8 hours per day, 240 days per year over a 45-year working life.
The objective of this exposure-response analysis was to compute the carcinogenic potency, expressed as the TC01, or the concentration of butadiene associated with a 1% excess probability of dying from leukemia. This analysis involved two stages. First, the relationship between exposure and the death rate due to leukemia within the cohort was modelled. This was accomplished by collapsing (or stratifying) the data into discrete exposure categories and then modelling the mean exposure in each category versus the death rates due to leukemia. In the second stage of analysis, the TC01 was calculated based on this exposure-response relationship and the background mortality rates in the Canadian population.
Exposure-response modelling:
In addition to stratifying by exposure, the data were stratified by race, age, calendar year, years since hire and styrene exposure in order to incorporate this information into the exposure-response relationship. Each of these variables was collapsed into a small number of discrete categories in order to reduce the number of strata, thereby improving model stability. These variables and their categories are presented in Table 7. Exposure, defined as the mean cumulative exposure per person-year, was calculated for person-years falling into each possible combination of the stratification variables.
The data were imported to Epicure (1993)7 for exposure-response modelling. All fitted models were of the form:
RR = O/E = g(D(t))
where RR is the rate ratio, O and E are the observed and expected numbers of leukemia deaths, D( t ) is cumulative butadiene exposure up to time t, and g is the exposure-response model, which is constrained to pass through one at zero exposure. Four different models, discussed in more detail below, were fitted to the data. At the model-fitting stage, the expected number of deaths is calculated on the basis of the nonexposed person-years in the cohort, and not from Canadian population background rates.
Lifetime probability of death due to leukemia:
Once the fitted exposure-response model was obtained, the lifetime probability of death due to leukemia was computed using lifetable methods taking into account the death rates in the Canadian population. The derivation of the formula used for the lifetime probability of death due to leukemia proceeds as follows.
Let d( t ) represent the exposure concentration of butadiene in ppm at age t years, and let D( t ) denote the cumulative exposure in ppm-years with:
This formulation of cumulative exposure allows for the possibility of non-constant exposure scenarios.
At a cumulative exposure of D( t ) ppmyears, the probability of dying from leukemia by age t is given by:
where hRD(t);t) is the mortality rate from leukemia at age t given a cumulative exposure to butadiene of D( t ), and S( t ) is the probability of survival up to age t. Equation (1) follows from the argument that the probability of death by age t is equal to the probability of death at age t multiplied by the probability of surviving up until age t. In lifetable analysis, the mortality and survival rates are constant for each year, so the integral in (1) can be replaced by a summation over year.
Exposure to butadiene is assumed to augment the background rate of leukemia for the Canadian population in a multiplicative fashion. In other words, the mortality rate, given exposure to butadiene, is equal to the background exposure rate multiplied by the excess risk due to exposure to butadiene. This is known as the "proportional hazard" model and is expressed as:
where h(t) is the background mortality rate from leukemia in the Canadian population, calculated from Canadian age-specific death rates8 due to leukemia, and g(D(t)) is the fitted exposure-response model, or excess risk at age t.
The survival rate, S(t), appearing in equation (1) is computed from Canadian agespecific death rates due to all causes, where the reported Canadian leukemia mortality rate is replaced by the modelled rate in order to incorporate exposure to butadiene. The formula describing the probability of survival up to age i is given by:
(3)
where hj * and hj are the Canadian mortality rates due to all causes and due to leukemia at age j, respectively, and gj = g(D(j)) is the excess risk at age j.
Substituting equation (2) into (1), the lifetime probability of death due to leukemia is given by:
where 1-70 years is the standard lifetime for a human.
Cancer potency (TC01):
The TC01 is computed by determining the exposure D(t) at which the excess risk is equal to 0.01. That is,
| P (D(t);t) - P(0;t)
_____________ 1 - P(0;t) |
= 0.01 |
If a constant exposure d is assumed for an individual from birth to age 70 years, then d(t) = d ppm and the cumulative exposure D(t) = d·t ppm-years. The TC01 is then the ambient exposure level d (in ppm) at which the excess risk equals 0.01 at t = 70 years.
Lagged exposure analysis:
In separate analyses, exposures were lagged by n = 2, 5, 10, 15, 20 and 25 years to determine if the models would provide better fits if the most recent n years of exposure were ignored. An n-year lag was achieved by resetting an individual's cumulative exposure at each year to be equal to the exposure he had accumulated n years prior. In so doing, the last n years of exposure do not affect the probability of developing leukemia. The data were first stratified on unlagged cumulative exposure, and then the individual exposures were lagged. Thus, the number of strata remains constant when using different lag periods, and models with different lags may be directly compared (Preston et al., 1987).
Validation study:
To assess the predictive power of the exposure- response models, a validation study was performed in which individuals in the cohort were divided randomly into two groups. The models were fit separately to both groups, and then a likelihood ratio test was performed to determine if the estimated parameters were equal. The process of dividing and fitting was repeated 1000 times to characterize the variability due to the random splitting process. If the models provided consistent fits, then the likelihood ratio test would be expected to reject at a rate equal to the desired significance level of the test (i.e., at a significance level of 0.05, the fitted parameters should be significantly different 1 in 20 times). If the tests are significant more often than this, the confidence in the predictive power of the models is reduced.
Figure 2 Observed rate ratios and fitted curves for leukemia in Delzell et al. (1995) study
Mean cumulative butadiene exposure per person-year (ppm-years)
Adjusted for age, calendar period, race, years since hire and styrene exposure.
Results
Exposure-response modelling:
Four different exposure-response models were examined and are presented in Table 8. These models are identical to those fitted in the Delzell et al. (1995) report except that model 2 is more general and flexible than the square root model used by those authors. Preliminary analysis indicated that all stratification variables except race significantly affected the model fit. Since race was only marginally insignificant, all variables were used to stratify the data prior to model fitting.
The four models were fitted while stratifying on race, age, calendar year, years since hire and styrene exposure. The results of the model fitting are displayed in Table 8. (N.B.: A smaller deviance roughly indicates a better fit.) A graphic representation of the data and the fitted models is shown in Figure 2. Judging from the model deviances and the shape of the curves relative to the data, especially in the low-dose region, model 1 provides the best fit to the data.
For purposes of comparison, the same models were fitted using the median exposure as per the Delzell et al. (1995) report. These analyses indicated that there is little difference between using mean or median exposures. Models including age as a multiplying factor of eγ·age instead of as a stratification variable were also fitted, but these models did not fit as well. Since cumulative exposure and years since hire may be confounded, their interaction was examined. The interaction was not significant for any of the models. The same models were refitted excluding the largest exposure group (200+ ppmyears), but this did not significantly affect any of the parameter estimates. The four models were also refitted allowing for different background rates for control and exposed populations. Different background rates might be necessary in occupational studies where lifetime nonexposed workers may differ fundamentally from exposed workers as a result of differences in jobs and work areas. Results of this analysis indicated that different background rates are not necessary for these data.
Table 8 Parameter estimates and model deviances for each of four models fitted to mean cumulative exposure per person-year for Delzell et al. (1995) study and comparison to parameter estimates from Delzell et al. analyses
The parameter estimates obtained in the present analysis are also not significantly different from those presented in the Delzell et al. (1995) report. The differences in parameter estimates are likely due to the different levels used in the stratification variables. Table 8 compares the parameter estimates obtained in this analysis with those of the Delzell et al. (1995) report.
Cancer potency (TC01):
The TC01s were calculated for each model using the lifetable methods described above, and the resulting ambient occupational exposures per person-year were converted to environmental exposures by assuming that the exposures occurred for 8 hours per day, 240 days per year. This amounts to multiplying the TC01 by:
| 8 hours _________ 24 hours |
× |
240 days _________ 365 days |
To convert the ambient exposures from ppm to mg/m3, the TC01s are further multiplied by 2.21, the conversion factor for butadiene. The occupational and equivalent environmental TC01s are presented in Table 9. Environmental TC01s for each of the four models ranged from 1.4 to 4.3 mg/m3. TC01s calculated excluding the largest exposure group were slightly smaller, ranging from 0.6 to 1.6 mg/m3, while those calculated on the basis of median exposures were similar, ranging from 0.4 to 5.0 mg/m3.
TC01s were also calculated using the parameter estimates from the Delzell et al. (1995) report and are compared with the TC01s developed here in Table 9. They ranged from 3.1 to 14.3 mg/m3.
Lagged exposure analysis:
The same four models were fitted when exposures were lagged by 2, 5, 10, 15, 20 and 25 years. The resulting model fits are displayed in Table 10. Since the deviances are similar for each lag period, this analysis indicates that lagging exposures does not dramatically improve the fit of any of the four models. In fact, TC01s for all four models and all lag periods ranged from 0.8 to 4.3 mg/m3.
Validation study:
With respect to model validation, the p-values for the tests of equality of the parameters are displayed in Table 11. If the models were providing consistent fits between the two halves, the proportion of p-values less than the significance level of α would be α . The results of the simulation study indicate that the test is rejecting more often than would be expected if the models were providing the same fits to both halves of the data. For model 1, the test was rejected at a significance level of 1% in 7.4% of the runs, whereas a rejection rate of 1% of the runs would be expected if the model was fitting consistently. The results of this analysis reduce the confidence in the power of the models to predict leukemia mortality rates.
Summary
It is noteworthy that the choice of the exposure-response model does not have a large impact on the resulting TC01; as indicated in Table 9, the values are similar, ranging from 1.4 to 4.3 mg/m3. However, if a best model must be chosen, it would be model 1, owing to the smaller deviance (Table 8), the shape of the curve relative to the data in the low-dose region (Figure 2) and the fact that it has one fewer parameter than model 2, which provides a similar fit. The TC01 for model 1 is 1.7 mg/m3.
It is difficult, though, to assess how well any of these models truly describes the data. It is noted that the plot in Figure 2 provides only a rough indication of the shape of the data, since each point on the plot is an average of data in many strata. The results of the validation study reduce confidence in the ability of the models to predict leukemia mortality.
The choice of exposure lag does not greatly improve the fit of any of the four models, and it does not affect the resulting TC01. Including all lagged models, the range of TC01s is still from 0.8 to 4.3 mg/m3.
For comparison with these values, potency estimates were also calculated on the basis of the recent case-control study of styrenebutadiene rubber workers by Matanoski et al. (1997). Although workers were from plants subsumed in the Delzell et al. (1995) study, exposure was independently characterized. Treating the odds ratio presented by these authors as a rate ratio (since leukemia is a rare disease) and using their model and parameter estimates as well as the same lifetable methods (1997). Although workers were from plants subsumed in the Delzell et al. (1995) study, exposure was independently characterized. Treating the odds ratio presented by these authors as a rate ratio (since leukemia is a rare disease) and using their model and parameter estimates as well as the same lifetable methods described above, the TC01 for environmental exposure was calculated to be 0.4 mg/m3. It is reassuring, therefore, that this value is only slightly lower than the estimates derived on the basis of the Delzell et al. (1995) cohort study data.
| Model | Proportion of p-values1 | ||
|---|---|---|---|
| less than 0.01 | less than 0.05 | les than 0.1 | |
| 1) RR = (1 + dose)α | 0.074 | 0.167 | 0.252 |
| 2) RR = 1 + β .doseα | 0.084 | 0.19 | 0.286 |
| 3) RR = eβ.dose | 0.08 | 0.188 | 0.264 |
| 4) RR = 1 + β .dose | 0.103 | 0.214 | 0.303 |
1 p-value of likelihood ratio test of equality of parameters fitted to each half of the data.
4 The potency estimate for carcinogenicity adopted in the Priority Substances Program is determined by calculating the dose or concentration associated with an increase in cancer incidence or mortality of an appropriate percentage. When based on toxicological data from studies in experimental animals, a 5% increase is generally chosen, as these values usually lie within or close to the observable range (i.e., a TC05 is calculated). When epidemiological data form the basis for derivation of a tumorigenic concentration, the percent increase selected is that which falls within the area of the exposure-response curve that represents the majority of the observable data; this is often less than 5%. In the case of butadiene, the carcinogenic potency calculated on the basis of modelling of epidemiological data (as described herein) was considered to be best defined as a 1% increase in mortality due to leukemia (i.e., a TC01).
5 Similar to tumorigenic concentrations (TC05s), benchmark concentrations for non-cancer effects (or BMC05s), when based on data in experimental animals, represent the dose or concentration associated with a 5% increase in the incidence of an effect compared with controls.
6 The cooperation of the sponsors and researchers for the Delzell et al. (1995) study in the provision of these data is gratefully acknowledged.
7 Epicure is a collection of interactive programs used to fit models to epidemiological data. The specific program used to model the data for this cohort of styrene-butadiene rubber workers is called AMFIT, which is specially designed to model hazard functions for censored cohort survival data. The strength of Epicure lies in its ability to easily allow the background rate to depend on user-specified strata, such as age, calendar period and race.
8 Mortality data were provided to Health Canada by Statistics Canada. The cooperation of the registrars of vital statistics in the provinces and territories of Canada who make mortality data available to Statistics Canada under federal-provincial agreements is gratefully acknowledged.
