- Email: [email protected]
Dose–response implications of the University of Alabama study of lymphohematopoietic cancer among workers exposed to 1,3-butadiene and styrene in the synthetic rubber industry
Chemico-Biological Interactions 135– 136 (2001) 637– 651
www.elsevier.com/locate/chembiont
Dose–response implications of the University of Alabama study of lymphohematopoietic cancer among workers exposed to 1,3-butadiene and styrene in the synthetic rubber industry Robert L. Sielken Jr *, Ciriaco Valdez-Flores Sielken & Associates Consulting Inc., Suite 230, 3833 Texas A6enue, Bryan, TX 77802, USA
Abstract New quantitative cancer risk estimates for exposure to 1,3-butadiene are presented. These estimates are based on the most recent human epidemiologic data developed by Drs Delzell and Macaluso and their colleagues at the University of Alabama at Birmingham. The implications of Poisson regression analyses of the relative rate for leukemia are explored using their updated dose estimates and lymphohematopoietic cancer data. The Poisson regression model in these analyses has the same form as in the U.S. Environmental Protection Agency (EPA)’s draft risk assessment of 1,3-butadiene [U.S. Environmental Protection Agency, Health Risk Assessment of 1,3-Butadiene — External Review Draft, National Center for Environmental Assessment, Office of Research and Development, 63 Fed. Reg. 7167 (February 12, 1998) Publication NCEA-W-0267, Washington, 1998]. Consistent with the proposed cancer risk assessment guidelines of the EPA and the EPA’s draft risk assessment, the exploration includes the maximum likelihood estimate of the ‘effective concentration’ (EC01) corresponding to an extra risk of leukemia of 0.01 (1%) from a lifetime continuous exposure to 1,3-butadiene based on a linear dose– response model and the cumulative 1,3-butadiene dose metric (ppm-years). The incorporation of the most recent exposure estimates results in a 2.5-fold decrease in the estimates of leukemia risks computed by EPA. In addition, three changes proposed by the American Chemistry Council (formerly the Chemical Manufacturers Association) to the EPA’s Science Advisory Board (SAB) for EPA’s draft risk assessment of 1,3-butadiene are incorporated into the calculation. This results in approximately an additional fivefold decrease in the risk estimates of leukemia. The leukemia cancer risk estimates in the EPA’s draft risk assessment of 1,3-butadiene decrease * Corresponding author. Tel.: + 1-979-8465175; fax: +1-979-8462671. E-mail address: [email protected] (R.L. Sielken, Jr). 0009-2797/01/$ - see front matter © 2001 Elsevier Science Ireland Ltd. All rights reserved. PII: S 0 0 0 9 - 2 7 9 7 ( 0 1 ) 0 0 2 1 7 - 4
638 R.L. Sielken, Jr, C. Valdez-Flores / Chemico-Biological Interactions 135–136 (2001) 637–651
by approximately a factor of 13-fold when the updated epidemiologic data and the alternative numbers proposed by industry to the SAB are both incorporated. Specifically, the maximum likelihood estimate of the EC01 increases from EPA’s 1.2 ppm to 2.8 ppm on the basis of the updated epidemiologic data and increases further to 15.1 ppm when the CMA’s proposed changes are also incorporated. © 2001 Elsevier Science Ireland Ltd. All rights reserved. Keywords: 1,3-Butadiene; Epidemiology; Lymphohematopoietic Leukemia risk estimates; Cancer risk assessment
cancer;
Dose – response
model;
1. Introduction In their most recent health risk assessment of 1,3-butadiene, the U.S. Environmental Protection Agency (EPA) based its 1,3-butadiene cancer potency estimate [1] on a Poisson linear model reported by Delzell et al. [2,3]. According to the EPA, the epidemiology-based risk estimates based on the linear rate ratio from that Poisson regression model are preferred ‘‘because of the high uncertainty in extrapolating 1,3-butadiene cancer risks from rodents to humans and the existence of good-quality occupational epidemiology data with exposure measures’’ [1]. The estimates of occupational exposures to 1,3-butadiene in the synthetic rubber industry generated and used by Dr Delzell and her colleagues (mostly at the University of Alabama at Birmingham (UAB)) in their initial reports in 1995 and 1996 [2,3] were the best available at that time. Nevertheless, the investigators indicated that the exposure estimates were very uncertain and not completely validated. The authors also stated that the subjects in the cohort were possibly misclassified with respect to process group and with respect to exposure levels. In 2000, a team led by Dr Macaluso at UAB presented a revised set of exposure estimates for 1,3-butadiene, styrene and dimethyldithiocarbamate [4]. The revised set of exposure estimates is ‘‘equivalent to the original set of estimates in ranking individual employees according to cumulative exposure levels’’ of 1,3-butadiene and styrene, but the new exposure estimates are generally substantially greater than the exposures in the original set [4]. The Chemical Manufacturers Association (CMA), now the American Chemistry Council (abbreviation CMA retained for clarity) proposed to EPA and its Science Advisory Board (SAB) in April 1998 [5] that at least three modifications (hereafter referred to as CMA-proposed changes) be made to their quantitative cancer risk assessment. These three CMA-proposed changes affect how UAB’s Poisson regression model is used to calculate a unit cancer risk (cancer potency) for regulatory purposes. Specifically, the three CMA-proposed changes refer to how the mathematical calculation of the maximum likelihood estimate of the ‘effective concentration’ (EC01) corresponding to an extra risk of leukemia of 0.01 (1%) from a lifetime continuous exposure to 1,3-butadiene may be calculated from the epidemiologic data using UAB’s linear dose– response model and the cumulative 1,3-butadiene dose metric (ppm-years).
R.L. Sielken, Jr, C. Valdez-Flores / Chemico-Biological Interactions 135–136 (2001) 637–651 639
New cancer risk estimates are calculated herein. These estimates are based on the most recent epidemiologic data with their updated set of exposures to 1,3-butadiene and styrene. The estimates use the same methodology that was used by the EPA in 1998 [1], and are presented both before and after incorporating the three changes specifically proposed to the SAB by industry. While the three CMA-proposed changes are important, they do not address several larger issues that are the subject of ongoing and anticipated future research activities. Even if the calculation of a unit cancer risk incorporates the CMA-proposed changes described below, the resulting cancer potency characterizations should be open to modifications based on future research and new findings.
2. Use of CMA-proposed changes in calculations EPA used Delzell et al.’s [2,3] linear Poisson regression model to characterize 1,3-butadiene’s dose – response relationship in humans. The rate ratio (RR) is a linear function of dose (cumulative ppm-years). The estimated multiplier of dose in RR is the basis for EPA’s unit cancer risk calculation. EPA combined this multiplier and the general US background leukemia rates to estimate the lifetime extra unit cancer risk due to environmental exposures from ages 0 to 85 years with no consideration of latency. The multiplier (0.0068) used by EPA is taken directly from the original 1995 Delzell et al. report [2]. Because EPA did not have direct access to the raw data in UAB’s study, the Agency was unable to alter the methodology used by UAB to incorporate person years and dose values into the dose–response modeling. UAB used methodology that was appropriate for hazard assessment, but that methodology has limitations if it is used to make subsequent unit risk calculations. Throughout the 1995 UAB Poisson regression modeling process, the dose (cumulative 1,3-butadiene exposure in ppm-years) for all person years in the highest dose group (greater than 200 ppm-years) was assigned to a somewhat arbitrary value of 250 ppm-years rather than the average ppm-years in this group. Furthermore, in EPA’s unit cancer risk calculation, EPA assigned a lifetime to be 85 years rather than the regulatory standard of 70 years and assigned the inhalation rate for chronic exposures to be greater than the rate recommended in the EPA’s Exposure Factors Handbook [8]. These three assignments substantially impacted EPA’s calculation of the unit cancer risk for 1,3-butadiene. These three assignments are the subject of the three CMA-proposed changes explicitly considered herein. Because EPA’s risk estimates were based on the original 1995 UAB exposure estimates, the magnitude of the impact of the first of the three CMA-proposed changes is assessed in the following section using the original 1995 exposure estimates. The magnitude of the first CMA-proposed change’s impact is roughly twice as large when the risk estimates are based on the updated 2000 UAB exposure estimates.
640 R.L. Sielken, Jr, C. Valdez-Flores / Chemico-Biological Interactions 135–136 (2001) 637–651
The proportional impacts of the second and third CMA-proposed changes on the cancer risk estimates are the same regardless of the exposure estimates.
2.1. Characterizing the mean response rate for person years with more than 200 ppm-years by the mean 6alue of the dose rather than an arbitrary choice of 250 ppm-years When UAB performed their Poisson regression modeling of the epidemiologic data and estimated the rate ratio (RR) to be RR = 1 + 0.0068 × cumulative ppm-years they used the value of 250 ppm-years to represent the highest dose group (more than 200 ppm-years). When we redid the Poisson regression modeling using UAB’s form of the model, namely, Mean leukemia response rate=Constant Background Hazard Rate × Age Effect× Calendar-year Effect × Years-since-hire Effect × Linear Rate Ratio but with the somewhat arbitrary value of 250 ppm-years replaced by the average value (370 ppm-years) for the person years with at least 200 cumulative ppm-years, then the multiplier of the cumulative ppm-years in the rate ratio decreased by approximately 22.5%. This decrease corresponds to approximately a 22.5% reduction (i.e. a 1.3-fold decrease, 100%/[100% − 22.5%]= 1.3) in the estimated cancer potency. Poisson regression is modeling the mean leukemia response rate. Because the response rate in a person year is assumed to be linearly related to the cumulative ppm-years for that person year, the mean response rate in the highest dose group is proportional to the mean cumulative ppm-years in that dose group. Therefore, using the average ppm-year for the highest dose group is more appropriate than using an arbitrary value like 250 ppm-years. Although we contend it would be more mathematically correct to use the mean for all intervals, in this case using the mid-point instead of the average for the non-highest dose intervals of cumulative 1,3-butadiene exposure had very little effect on the estimates of the cancer potency.
2.2. Characterizing the excess cancer risk for a 70 -year lifetime In quantifying the potential excess cancer risks due to 1,3-butadiene exposure, EPA departed from their traditional default assumption of a 70-year lifetime and assumed an 85-year lifetime. According to the National Center for Health Statistics [9], a person born in the USA in 1990 is only expected to live 75.4 years (males 72.0 years and females 78.8 years). The quantitative impact of switching from the traditional
R.L. Sielken, Jr, C. Valdez-Flores / Chemico-Biological Interactions 135–136 (2001) 637–651 641
70-year lifetime to an assumed 85-year lifetime is to increase the risk characterization approximately threefold. Specifically, using EPA’s linear model for the rate ratio (1+ 0.0068× cumulative ppm-years), the added cancer risk associated with a continuous environmental exposure (24 h a day, every day of a lifetime) to 0.0001 ppm 1,3-butadiene is 0.88 ×10 − 6 (0.88 in one million) for 85 years at risk (age 0 to age 85) as opposed to only 0.32× 10 − 6 (0.32 in one million) for 70 years at risk (age 0 to age 70). Similarly, the added cancer risk is only 0.48× 10 − 6 (0.48 in one million) for 75 years at risk (age 0 to age 75). The risk for 75 years is 55% of the risk for 85 years. The risk for 70 years is only 36% of the risk for 85 years. By changing the standard lifetime from 70 years to 85 years, EPA has increased the risk characterization by 2.75-fold, as shown in Table 1.
2.3. Simplified explanation of the lifetime extra cancer risk calculation and its implications The following is a simplified explanation of the extra risk calculation. This explanation shows how the 0.0068 multiplier in the linear model for the rate ratio leads to a cancer potency of 0.0087 per ppm. The form of the calculation indicates why an increase from a 70-year lifetime to a longer lifetime results in a disproportionate increase in the lifetime extra unit cancer risk when the relevant dose metric is assumed to be cumulative ppm-years and it is assumed that all exposures before a specific age cause a multiplicative increase in the leukemia hazard rate for that age. In addition, the form of the calculation explains why the value of the lifetime excess cancer risk is substantially impacted by the background hazard rate. For simplicity, the adjustment for competing risks is omitted from the following explanation. The adjustment for competing risks (although somewhat complicated mathematically) was included (as it should have been) in the actual calculation of the cancer potency of 0.0087. Apart from the adjustment for competing risks, the lifetime extra cancer risk at c ppm is calculated as follows:
Table 1 Impact of the length of the lifetime on the estimation of the additional risks using EPA’s linear rate ratio Length of lifetime in years
Additional risk of leukemia at 0.0001 ppm
Percentage of additional risk for a lifetime of 85 years
70 75 80 85
0.3205×10−6 0.4826×10−6 0.6776×10−6 0.8811×10−6
36% 55% 77% 100%
(100%/2.75) (100%/1.83) (100%/1.30) (100%/1.00)
642 R.L. Sielken, Jr, C. Valdez-Flores / Chemico-Biological Interactions 135–136 (2001) 637–651
Extra Cancer Risk =Background Rate for the Specified Response in Year 1× 0.0068× 1 year × c ppm +Background Rate for the Specified Response in Year 2 × 0.0068 × 2 years× c ppm… + Background Rate for the Specified Response in Year 70× 0.0068 ×70 years× c ppm… +Background Rate for the Specified Response in Year 85× 0.0068 ×85 years× c ppm and, approximately, Extra Cancer Risk= c ppm × 0.0087 Because the cumulative dose increases with age for constant environmental exposures, the multiplier (here, 0.0068× age) of the background hazard rate increases as the age increases. Also, for many responses (including leukemia) the background hazard rate increases with age. Thus, the contribution to the lifetime extra cancer risk increases in magnitude as the age increases. An increase from a 70-year lifetime to an 85-year lifetime results in a disproportionate increase in the lifetime extra cancer risk. The form of the calculation also indicates how the value of the cancer potency is strongly determined by the choice of the Background Hazard Rate and the value that is substituted for ‘Background Rate for Specified Response in Year – – – ’ in the calculation of the cancer potency. For example, if this rate is halved, then the cancer potency is halved.
2.4. Characterizing the excess cancer risk for en6ironmental exposures based on the EPA Exposure Factors Handbook’s 6alue for the a6erage amount of air inhaled per day In converting from occupational exposure (8 h per day, 240 days per year) to continuous environmental exposure (24 h per day, 365 days per year), EPA assumed that humans inhale 20 m3 per day. This assumption is equivalent to assuming that environmentally exposed humans are approximately three times more exposed than occupationally exposed humans, because (20 m3/10 m3) × (365 days/240 days)= 3.042. Here, ‘three times more exposed’ means, for example, environmental exposure to 1 ppm results in three times more 1,3-butadiene intake than occupational exposure to 1 ppm. In other words, 1 ppm-year of environmental exposure is approximately equivalent to 3 ppm-years of occupational exposure.
R.L. Sielken, Jr, C. Valdez-Flores / Chemico-Biological Interactions 135–136 (2001) 637–651 643
Fig. 1. Distribution of the cumulative dose (ppm-years) of the workers that died with leukemia based on the updated and original estimates of 1,3-butadiene exposure.
Combining the average adult inhalation rates for men and women listed in the 1989 EPA Exposure Factors Handbook [6], the American Industrial Health Council Exposure Factors Sourcebook [7] recommends that the ‘‘daily inhalation rate for adults for use in risk assessments is 18 m3/day.’’ The 1997 EPA Exposure Factors Handbook [8] recommends even smaller values (11.3 m3/day for females and 15.2 m3/day for males). If EPA assumed that humans inhale 18 m3 per day, then environmentally exposed humans would be approximately 2.7 times more exposed than occupationally exposed humans, because (18 m3/10 m3) × (365 days/240 days)= 2.7375 Using an inhalation rate of 18 m3 instead of 20 m3 would have decreased the cancer potency by 10% from 0.0087 per ppm to 0.0078 per ppm (i.e. a 1.11-fold decrease). Using 1997 EPA values would result in larger decreases. 3. Updated UAB exposure database and analyses In 2000, Macaluso et al. [4] updated their exposure estimates for 1,3-butadiene, styrene, and dimethyldithiocarbamate among synthetic rubber workers. At the same time, Delzell et al. [10] reported a new assessment of lymphohematopoietic cancer and exposure to 1,3-butadiene and other chemicals. Under an agreement with the International Institute of Synthetic Rubber Producers, the UAB provided the authors with electronic files containing the updated epidemiological database of worker exposure histories. According to Macaluso et al.’s
644 R.L. Sielken, Jr, C. Valdez-Flores / Chemico-Biological Interactions 135–136 (2001) 637–651
final report [4], the updated exposure estimates are considerably higher than the original exposure estimates derived in 1995. Specifically, according to Macaluso et al., the updated job- and subject-specific exposures to 1,3-butadiene were up to one order of magnitude higher than the 1995 estimates. However, despite the significant difference between the magnitude of the updated 1,3-butadiene exposure estimates and the magnitude of the original exposures, the ranking of individual employees according to cumulative exposure levels using the original and updated estimates were equivalent. Thus, Macaluso et al. state that risk assessment analyses based on the updated exposure estimates are expected to decrease considerably when compared to risk characterizations based on the original exposure estimates. Fig. 1 shows the distribution of the cumulative dose (ppm-years) of the workers that died with leukemia based on the updated and the original estimates of 1,3-butadiene exposure. The average cumulative 1,3-butadiene exposure for workers that died with leukemia is 430 ppm-years based on the updated exposure estimates compared with an average of 66 ppm-years based on the original exposure estimates. The average ppm-years for leukemia decedents based on the updated exposure estimates is more than sixfold greater than the average ppm-years computed using the original exposure estimates.
4. Calculation of effective concentrations with updated UAB exposures and CMA-proposed changes The EPA used UAB’s 1995 linear Poisson regression model results to calculate the maximum likelihood estimate of the ‘effective concentration’ (EC01) and the 95% lower confidence limit on the effective concentration (LEC01) associated with an extra risk of leukemia of 0.01 (1%) from a lifetime exposure. Herein, the same form of the linear Poisson regression model is used with the new worker exposure estimates derived by Macaluso et al. [4] to update the EC01 and LEC01 values. Thus, the dose– response methodology is consistent with EPA’s draft risk assessment of 1,3-butadiene and the proposed cancer risk assessment guidelines of the EPA. The impacts of updated exposure values on the estimates of EC01 and LEC01 are indicated as well as the impacts of the three CMA-proposed changes discussed above. The dose – response component of the default linear Poisson regression model used by EPA and used herein is RR = 1 + i × dose where RR is the rate ratio (rate of leukemia deaths per person year with exposure equal to ‘dose’ divided by the rate of leukemia deaths per person year with zero exposure), i is the slope of the linear rate ratio function, and ‘dose’ is the cumulative exposure to 1,3-butadiene in ppm-years. In both EPA’s assessment and the analyses herein, the complete Poisson model incorporates the effect of possible covariates as follows:
R.L. Sielken, Jr, C. Valdez-Flores / Chemico-Biological Interactions 135–136 (2001) 637–651 645
Mean Leukemia Response Rate=Background Rate× Effect of Age × Effect of Calendar Year × Effect of Years Since Hire × Effect of Styrene × Linear Rate Ratio for Leukemia
(1)
where the background rate is the average leukemia deaths per person-year when the effect of the covariates is null (i.e. excluding the age effect, excluding the calendar year effect, excluding the styrene exposure, etc.). As in EPA’s assessment, the age covariate has values for each of the following intervals [40, 50), [50, 60), [60, 70), [70, 80), and 80 or more years of age; the calendar year covariate has intervals [1950, 1960), [1960, 1970), [1970, 1980), [1980, 1990), and [1990, 1991] where 1991 is the end of the follow-up; and the intervals for the year-since-hire covariate are [10, 20), [20, 30), and 30 or more years since hire. For styrene, the intervals are based on the quintiles used by Delzell et al. [10] in the final report (i.e. intervals for the cumulative exposure to styrene are defined such that each interval includes approximately the same number of deaths with leukemia). These intervals for styrene are 0, (0, 10.4], (10.4, 28.3], (28.3, 40.6], (40.6, 98.1], and more than 98.1 ppm-years. All the intervals for the covariates, except for the intervals for styrene exposure, are identical to those used in the Delzell et al. 1995 report [2] on which EPA based its risk assessment. The intervals to control for the effect of styrene exposures are constructed in the same way as in the 1995 report but are adjusted to reflect the updated exposure estimates reported by Macaluso et al. [4]. The cumulative 1,3-butadiene exposure in ppm-years is partitioned into intervals for the estimation of the i parameter in the linear rate ratio for leukemia. The intervals were defined so that each group of person years includes approximately the same number of leukemia deaths. The intervals are 0, (0, 35.6], (35.6, 69.3], (69.3, 180.5], (180.5, 265.6], (265.6, 400.9], (400.9, 791.9], and more than 791.9 ppm-years. Table 2 contains the cumulative 1,3-butadiene exposure intervals along with the observed number of leukemia deaths and person years included in each dose interval. In the evaluation of the linear rate ratio for leukemia, the average dose for each group is assumed to be equal to the mid-point of the interval (consistent with EPA’s draft risk assessment), except for the last group. The average dose for the last group is computed according to the CMA-proposed change and is equal to the average cumulative 1,3-butadiene exposure over all the person years with cumulative 1,3-butadiene greater than 791.9 ppm-years. In the maximum likelihood estimation for the Poisson regression model, the observation period for each of the 13 131 individuals in the analysis is subdivided into person years. The 234 430.6 person years are partitioned according to the age, calendar year, years since hire, cumulative styrene exposure, and cumulative 1,3-butadiene exposure intervals. (Cumulative exposures are computed from the beginning of exposure to the end of the person year.) There are 3600 partitions (five age intervals, five calendar year intervals, three years-since-hire intervals, six
Dose group (ppm-years)
0 (0, 35.6] (35.6, 69.3] (69.3, 180.5] (180.5, 265.6] (265.6, 400.9] (400.9, 791.9] More than 791.9 Summary
Point for parametric estimation
0.00 17.80 52.45 124.90 223.05 333.25 596.40 1776.71
Observed number of leukemia deaths
Number of person years in the group
Leukemia death rate per 10 000 person years Obs.
Pred.
7 8 7 7 7 8 7 8
47 911.2 59 206.4 27 923.1 43 414.2 15 655.1 14 648.9 14 961.3 10 710.4
1.46 1.35 2.51 1.61 4.47 5.46 4.68 7.47
1.46 1.47 1.90 2.69 3.19 3.82 4.41 8.59
59
234 430.6
2.52
2.52
646 R.L. Sielken, Jr, C. Valdez-Flores / Chemico-Biological Interactions 135–136 (2001) 637–651
Table 2 Summary of the cumulative 1,3-butadiene exposure (ppm-years) intervals used in the maximum likelihood estimation and the observed and predicted leukemia death rates
R.L. Sielken, Jr, C. Valdez-Flores / Chemico-Biological Interactions 135-136 (2001) 637–651 647
cumulative styrene exposure intervals, and eight cumulative 1,3-butadiene exposure intervals; 5× 5 ×3 ×6 × 8 =3600). The observed data for the maximum likelihood estimation are, for each of the 3600 partitions, the number of person years and the number of those years in which a leukemia death occurred. (Table 2 indicates a summary of the observed data (the sums of the values for the 450 partitions for each of the eight cumulative 1,3-butadiene exposure intervals) but does not indicate the values for each of the 3600 partitions individually and explicitly considered in the maximum likelihood estimation.) The likelihood is the product over the partitions of the Poisson probability of the observed number of leukemia deaths in the partition; namely, u xe − u/x! where x is the observed number of leukemia deaths in the partition and u is the mean response rate for the partition (i.e. the number of person years in the partition multiplied by the mean leukemia response rate (Eq. (1)) for a person year in the partition). The mean leukemia response rate (Eq. (1)) for a person year in a partition contains a parameter for the background rate, the age interval, the calendar-year interval, the years-since-hire interval, the styrene exposure interval, and the parameter in the linear rate ratio for leukemia. In the maximum likelihood estimation, 17 parameters are estimated; specifically, one parameter for the background rate, four parameters for the five age intervals, four parameters for the five calendar-year intervals, two parameters for the three years-since-hire intervals, five Table 3 Parameters in the Poisson regression model and their maximum likelihood estimates Multiplicative effect
Interval
Background rate
0.00006
Age (years)
[40, 50) [50, 60) [60, 70) [70, 80) 80 or more
Calendar year
[1950, [1960, [1970, [1980, [1990,
Years since hire
[10, 20) [20, 30) 30 or more
Styrene exposure (ppm-years)
Parameter value
1960) 1970) 1980) 1990) 1991]
0 (0, 10.4] (10.4, 28.3] (28.3, 40.6] (40.6, 98.1] More than 98.1 i: multiplier of cumulative 1,3-butadiene exposure in linear rate ratio for leukemia
1.0 1.57854 1.95228 5.67143 11.68332 1.0 0.33776 0.61959 0.60573 1.35311 1.0 2.72331 1.26667 1.0 0.91376 1.47260 3.07787 1.50670 1.54735 0.00157
648 R.L. Sielken, Jr, C. Valdez-Flores / Chemico-Biological Interactions 135–136 (2001) 637–651
Fig. 2. Comparison of the observed and predicted leukemia death rates per 10 000 person years.
parameters for the six styrene intervals, and one i parameter. (Because the model contains a background rate, the multiplicative effect of each covariate is, without loss of generality, normalized to 1.0 for the first interval and is an estimated parameter value for each remaining interval.) The maximum likelihood estimates of the parameters are in Table 3. A comparison of the observed and predicted leukemia response rates (summarized for each cumulative 1,3-butadiene exposure interval) is given in Table 2 and Fig. 2. The maximum likelihood estimate of the linear rate ratio using the Poisson regression model described above is RR =1 + 0.0016 × Cumulative 1,3-Butadiene Exposure (ppm-years) The maximum likelihood estimate of the rate ratio is used to estimate the lifetime excess risk of leukemia mortality for specified levels of continuous exposures to 1,3-butadiene. These lifetime risks are computed using the relative rate estimates and an actuarial approach that takes into account the 1990 U.S. background rate of leukemia deaths and the effects of competing causes of death reported in the 1990 U.S. population census [9]. Environmental exposures to 1,3-butadiene are assumed to be continuous during the entire lifetime. The lifetime of 70 years suggested as one of the CMA-proposed changes is used in this risk characterization. The occupational 1,3-butadiene exposures in the UAB epidemiology study are converted to continuous environmental exposures by multiplying the occupational exposure estimates by a factor that accounts for the volume of air inhaled per day (10 m3 in the workplace/18 m3 in the environment) and by another factor that accounts for the number of days exposed per year (240 days in the workplace/365 days in the environment).
R.L. Sielken, Jr, C. Valdez-Flores / Chemico-Biological Interactions 135–136 (2001) 637–651 649
The Poisson regression model estimate of the effective concentration EC01 satisfies 0.01 = [P(EC01) − P(0)]/[1 − P(0)] where P(EC01) is the probability of leukemia death by age 70 if continuously exposed to EC01 ppm of 1,3-butadiene. The actuarial approach for the computation of P(EC01) is based on P(EC01) =Sum over the years (t) from birth to 70 {[1990 U.S. Background Leukemia Death Rate at Age t] × [Rate Ratio Evaluated for Cumulative Dose at EC01] × [U.S. Probability of not Dying from a Competing Cause of Death Before Age t]}
The best fitting linear Poisson regression model implies that the maximum likelihood estimate of the effective concentration EC01 corresponding to an extra risk of 0.01 (1%) from a lifetime of continuous exposure to 1,3-butadiene is EC01 =15.1 ppm The equivalent cancer potency based on the default procedure of linearly extrapolating the extra risk of 0.01 at EC01 back to zero extra risk at dose zero is Unit Risk= 0.01/EC01 =0.01/15.1 =0.00066 per ppm The new EPA guidelines emphasize the use of effective concentrations as the points of departure for low-dose risk characterization. The effective concentration corresponding to an excess risk of 0.10 (10%) may be reasonable for animal studies, but it is not reasonable for human epidemiologic studies. Workers in epidemiologic studies usually experience exposure levels that are well below the EC10. Thus, for human epidemiologic studies, excess risks of 0.01 or below are more appropriate than 0.10. Furthermore, the estimated exposure levels corresponding to a 0.01 excess risk are more consistent and less variable than the estimated exposure levels associated with 0.10 excess risk with respect to alternative forms of the Poisson regression model (i.e. alternative forms of the relative ratio function of the cumulative exposure). EPA used the 0.01 excess risk level to compute the effective concentration in their draft risk assessment. The 95% lower confidence limit (LEC01) on the EC01 is computed using the 95% upper confidence limit on the slope parameter in the linear rate ratio. The 95% upper confidence limit on the slope is computed using EPA’s default procedure that is based on the likelihood ratio technique. The resulting LEC01 is equal to LEC01 =9.0 ppm with the corresponding upper bound on cancer potency equal to Upper Bound on Unit Risk= 0.01/9.0 = 0.0011 per ppm
650 R.L. Sielken, Jr, C. Valdez-Flores / Chemico-Biological Interactions 135–136 (2001) 637–651
The effective dose EC01 of 15.1 ppm includes the changes proposed to the SAB and EPA by the CMA. If the CMA-proposed changes had not been included, then the estimated EC01 would have been 2.8 ppm. Overall, the impact of using the three CMA-proposed changes is a 5.4-fold increase in the EC01 from 2.8 to 15.1 ppm and a 5.4-fold decrease in the maximum likelihood estimate of the unit risk from 0.0036 to 0.00066 per ppm. The first CMA-proposed change (namely, the use of the average cumulative exposure for the last dose interval) increases the EC01 by approximately 1.76-fold. This increase of 1.76-fold in the EC01 using the updated exposure estimates is greater than the increase of 1.30-fold when the original exposure estimates were used. The other two CMA-proposed changes (namely, the use of a daily inhalation rate of 18 m3 and a lifetime of 70 years) increase the EC01 by approximately 1.11and 2.74-fold, respectively, as before.
5. Comparison of new and previous cancer risk characterizations New leukemia risk characterizations have been developed using a methodology that is consistent with EPA’s draft risk assessment of 1,3-butadiene and EPA’s proposed cancer risk assessment guidelines. The new risk characterizations use the updated exposure estimates to 1,3-butadiene and incorporate the changes proposed by the CMA for the EPA’s draft risk assessment of 1,3-butadiene. The risk characterization published by EPA in their draft risk assessment of 1,3-butadiene estimated an EC01 equal to 1.16 ppm (or, equivalently, a cancer potency of 0.0087 per ppm). The risk characterization with updated exposure concentrations estimates an EC01 equal to 2.8 ppm (or a cancer potency of 0.0036 per ppm). That is, the incorporation of the updated exposure values in the estimation of linear Poisson regression model increases the EC01 (i.e. reduces the estimated cancer potency of 1,3-butadiene) by approximately a factor of 2.4. Incorporating the CMA-proposed changes increases the EC01 from 2.8 ppm to 15.1 ppm. This is an additional 5.4-fold increase over the estimated EC01 before the CMA-proposed changes are included. Overall, the EC01 (15.1 ppm) estimated in the present risk characterization is approximately 13-fold greater than the EC01 (1.16 ppm) estimated in the risk characterization presented by EPA in their draft risk assessment of 1,3-butadiene. The potency value of 0.0087 per ppm estimated with the original exposure data is more than 13-fold greater than the potency value of 0.00066 per ppm estimated by us using the updated, more reliable, exposure estimates and the CMA-proposed changes.
6. Conclusions and future research In 1998 EPA’s draft risk assessment for 1,3-butadiene estimated an EC01 equal to 1.16 ppm and a cancer potency of 0.0087 per ppm based on UAB’s human epidemiologic data. Using EPA’s methodology, UAB’s updated exposure estimates,
R.L. Sielken, Jr, C. Valdez-Flores / Chemico-Biological Interactions 135–136 (2001) 637–651 651
and the three CM-proposed changes, the estimated EC01 is 15.1 ppm, and the cancer potency is 0.00066 per ppm. UAB’s updated exposure estimates, which have been extensively reviewed, are a significant step in the direction of more reliable estimation. The reliability of risk characterizations of 1,3-butadiene using epidemiologic data, however, can still be improved. Currently, the cohort includes exposures and follows workers through the end of 1991. The cohort can be followed for a longer period to include more accurate estimation in the older years of life. Although peak 1,3-butadiene exposures and average 1,3-butadiene exposures for concentrations above and below 100 ppm have been estimated, the role of these exposures in lymphohematopoietic cancers and cancer dose–response modeling needs to be clarified. The most biologically relevant dose metric (e.g. combination of the duration and intensity of exposure, toxicokinetic and toxicodynamic doses) for dose–response modeling and risk characterization needs to be determined. Similarly, the effect of a lag in exposures and the effect of latency need to be further investigated. References [1] U.S. Environmental Protection Agency, Health Risk Assessment of 1,3-Butadiene — External Review Draft, National Center for Environmental Assessment, Office of Research and Development, 63 Fed. Reg. 7167 (February 12, 1998) Publication NCEA-W-0267, Washington, 1998. [2] E. Delzell, N. Sathiakumar, M. Macaluso, M. Hovinga, R. Larson, F. Barbone, C. Beall, P. Cole, J. Julian, D.C.F. Muir, A follow-up study of synthetic rubber workers: A report submitted to the International Institute of Synthetic Rubber Producers, 1995. [3] E. Delzell, N. Sathiakumar, M. Hovinga, M. Macaluso, J. Julian, R. Larson, P. Cole, D. Muir, A follow-up study of synthetic rubber workers, Toxicology 113 (1996) 182 – 189. [4] M. Macaluso, R. Larson, J. Lynch, S. Lipton, E. Delzell, Historical estimation of exposure to 1,3-butadiene, styrene and dimethyldithiocarbamate among synthetic rubber workers: Final report submitted to the International Institute of Synthetic Rubber Producers, 2000. [5] Chemical Manufacturers Association, Comments of the olefins panel of the Chemical Manufacturers Association on EPA’s Draft Health Risk Assessment for 1,3-Butadiene, 1998. [6] U.S. Environmental Protection Agency, Office of Health and Environmental Assessment, Exposure Factors Handbook, EPA/600/08-89/043, Washington, 1989. [7] American Industrial Health Council (AIHC), Exposure Factors Sourcebook, AIHC, Washington, 1994. [8] U.S. Environmental Protection Agency, Office of Research and Development, Exposure factors handbook, Volume I — General Factors, EPA/600/P-95/002Fa, Washington, 1997. [9] U.S. Department of Health and Human Services. Vital Statistics of the United States 1990, Volume II — Mortality, Part A. [10] E. Delzell, M. Macaluso, N. Sathiakumar, R. Matthews, Lymphohematopoietic cancer among workers exposed to 1,3-butadiene, styrene and dimethyldithiocarbamate in the synthetic rubber industry, Final report submitted to the International Institute of Synthetic Rubber Producers, 2000.
.
www.elsevier.com/locate/chembiont
Dose–response implications of the University of Alabama study of lymphohematopoietic cancer among workers exposed to 1,3-butadiene and styrene in the synthetic rubber industry Robert L. Sielken Jr *, Ciriaco Valdez-Flores Sielken & Associates Consulting Inc., Suite 230, 3833 Texas A6enue, Bryan, TX 77802, USA
Abstract New quantitative cancer risk estimates for exposure to 1,3-butadiene are presented. These estimates are based on the most recent human epidemiologic data developed by Drs Delzell and Macaluso and their colleagues at the University of Alabama at Birmingham. The implications of Poisson regression analyses of the relative rate for leukemia are explored using their updated dose estimates and lymphohematopoietic cancer data. The Poisson regression model in these analyses has the same form as in the U.S. Environmental Protection Agency (EPA)’s draft risk assessment of 1,3-butadiene [U.S. Environmental Protection Agency, Health Risk Assessment of 1,3-Butadiene — External Review Draft, National Center for Environmental Assessment, Office of Research and Development, 63 Fed. Reg. 7167 (February 12, 1998) Publication NCEA-W-0267, Washington, 1998]. Consistent with the proposed cancer risk assessment guidelines of the EPA and the EPA’s draft risk assessment, the exploration includes the maximum likelihood estimate of the ‘effective concentration’ (EC01) corresponding to an extra risk of leukemia of 0.01 (1%) from a lifetime continuous exposure to 1,3-butadiene based on a linear dose– response model and the cumulative 1,3-butadiene dose metric (ppm-years). The incorporation of the most recent exposure estimates results in a 2.5-fold decrease in the estimates of leukemia risks computed by EPA. In addition, three changes proposed by the American Chemistry Council (formerly the Chemical Manufacturers Association) to the EPA’s Science Advisory Board (SAB) for EPA’s draft risk assessment of 1,3-butadiene are incorporated into the calculation. This results in approximately an additional fivefold decrease in the risk estimates of leukemia. The leukemia cancer risk estimates in the EPA’s draft risk assessment of 1,3-butadiene decrease * Corresponding author. Tel.: + 1-979-8465175; fax: +1-979-8462671. E-mail address: [email protected] (R.L. Sielken, Jr). 0009-2797/01/$ - see front matter © 2001 Elsevier Science Ireland Ltd. All rights reserved. PII: S 0 0 0 9 - 2 7 9 7 ( 0 1 ) 0 0 2 1 7 - 4
638 R.L. Sielken, Jr, C. Valdez-Flores / Chemico-Biological Interactions 135–136 (2001) 637–651
by approximately a factor of 13-fold when the updated epidemiologic data and the alternative numbers proposed by industry to the SAB are both incorporated. Specifically, the maximum likelihood estimate of the EC01 increases from EPA’s 1.2 ppm to 2.8 ppm on the basis of the updated epidemiologic data and increases further to 15.1 ppm when the CMA’s proposed changes are also incorporated. © 2001 Elsevier Science Ireland Ltd. All rights reserved. Keywords: 1,3-Butadiene; Epidemiology; Lymphohematopoietic Leukemia risk estimates; Cancer risk assessment
cancer;
Dose – response
model;
1. Introduction In their most recent health risk assessment of 1,3-butadiene, the U.S. Environmental Protection Agency (EPA) based its 1,3-butadiene cancer potency estimate [1] on a Poisson linear model reported by Delzell et al. [2,3]. According to the EPA, the epidemiology-based risk estimates based on the linear rate ratio from that Poisson regression model are preferred ‘‘because of the high uncertainty in extrapolating 1,3-butadiene cancer risks from rodents to humans and the existence of good-quality occupational epidemiology data with exposure measures’’ [1]. The estimates of occupational exposures to 1,3-butadiene in the synthetic rubber industry generated and used by Dr Delzell and her colleagues (mostly at the University of Alabama at Birmingham (UAB)) in their initial reports in 1995 and 1996 [2,3] were the best available at that time. Nevertheless, the investigators indicated that the exposure estimates were very uncertain and not completely validated. The authors also stated that the subjects in the cohort were possibly misclassified with respect to process group and with respect to exposure levels. In 2000, a team led by Dr Macaluso at UAB presented a revised set of exposure estimates for 1,3-butadiene, styrene and dimethyldithiocarbamate [4]. The revised set of exposure estimates is ‘‘equivalent to the original set of estimates in ranking individual employees according to cumulative exposure levels’’ of 1,3-butadiene and styrene, but the new exposure estimates are generally substantially greater than the exposures in the original set [4]. The Chemical Manufacturers Association (CMA), now the American Chemistry Council (abbreviation CMA retained for clarity) proposed to EPA and its Science Advisory Board (SAB) in April 1998 [5] that at least three modifications (hereafter referred to as CMA-proposed changes) be made to their quantitative cancer risk assessment. These three CMA-proposed changes affect how UAB’s Poisson regression model is used to calculate a unit cancer risk (cancer potency) for regulatory purposes. Specifically, the three CMA-proposed changes refer to how the mathematical calculation of the maximum likelihood estimate of the ‘effective concentration’ (EC01) corresponding to an extra risk of leukemia of 0.01 (1%) from a lifetime continuous exposure to 1,3-butadiene may be calculated from the epidemiologic data using UAB’s linear dose– response model and the cumulative 1,3-butadiene dose metric (ppm-years).
R.L. Sielken, Jr, C. Valdez-Flores / Chemico-Biological Interactions 135–136 (2001) 637–651 639
New cancer risk estimates are calculated herein. These estimates are based on the most recent epidemiologic data with their updated set of exposures to 1,3-butadiene and styrene. The estimates use the same methodology that was used by the EPA in 1998 [1], and are presented both before and after incorporating the three changes specifically proposed to the SAB by industry. While the three CMA-proposed changes are important, they do not address several larger issues that are the subject of ongoing and anticipated future research activities. Even if the calculation of a unit cancer risk incorporates the CMA-proposed changes described below, the resulting cancer potency characterizations should be open to modifications based on future research and new findings.
2. Use of CMA-proposed changes in calculations EPA used Delzell et al.’s [2,3] linear Poisson regression model to characterize 1,3-butadiene’s dose – response relationship in humans. The rate ratio (RR) is a linear function of dose (cumulative ppm-years). The estimated multiplier of dose in RR is the basis for EPA’s unit cancer risk calculation. EPA combined this multiplier and the general US background leukemia rates to estimate the lifetime extra unit cancer risk due to environmental exposures from ages 0 to 85 years with no consideration of latency. The multiplier (0.0068) used by EPA is taken directly from the original 1995 Delzell et al. report [2]. Because EPA did not have direct access to the raw data in UAB’s study, the Agency was unable to alter the methodology used by UAB to incorporate person years and dose values into the dose–response modeling. UAB used methodology that was appropriate for hazard assessment, but that methodology has limitations if it is used to make subsequent unit risk calculations. Throughout the 1995 UAB Poisson regression modeling process, the dose (cumulative 1,3-butadiene exposure in ppm-years) for all person years in the highest dose group (greater than 200 ppm-years) was assigned to a somewhat arbitrary value of 250 ppm-years rather than the average ppm-years in this group. Furthermore, in EPA’s unit cancer risk calculation, EPA assigned a lifetime to be 85 years rather than the regulatory standard of 70 years and assigned the inhalation rate for chronic exposures to be greater than the rate recommended in the EPA’s Exposure Factors Handbook [8]. These three assignments substantially impacted EPA’s calculation of the unit cancer risk for 1,3-butadiene. These three assignments are the subject of the three CMA-proposed changes explicitly considered herein. Because EPA’s risk estimates were based on the original 1995 UAB exposure estimates, the magnitude of the impact of the first of the three CMA-proposed changes is assessed in the following section using the original 1995 exposure estimates. The magnitude of the first CMA-proposed change’s impact is roughly twice as large when the risk estimates are based on the updated 2000 UAB exposure estimates.
640 R.L. Sielken, Jr, C. Valdez-Flores / Chemico-Biological Interactions 135–136 (2001) 637–651
The proportional impacts of the second and third CMA-proposed changes on the cancer risk estimates are the same regardless of the exposure estimates.
2.1. Characterizing the mean response rate for person years with more than 200 ppm-years by the mean 6alue of the dose rather than an arbitrary choice of 250 ppm-years When UAB performed their Poisson regression modeling of the epidemiologic data and estimated the rate ratio (RR) to be RR = 1 + 0.0068 × cumulative ppm-years they used the value of 250 ppm-years to represent the highest dose group (more than 200 ppm-years). When we redid the Poisson regression modeling using UAB’s form of the model, namely, Mean leukemia response rate=Constant Background Hazard Rate × Age Effect× Calendar-year Effect × Years-since-hire Effect × Linear Rate Ratio but with the somewhat arbitrary value of 250 ppm-years replaced by the average value (370 ppm-years) for the person years with at least 200 cumulative ppm-years, then the multiplier of the cumulative ppm-years in the rate ratio decreased by approximately 22.5%. This decrease corresponds to approximately a 22.5% reduction (i.e. a 1.3-fold decrease, 100%/[100% − 22.5%]= 1.3) in the estimated cancer potency. Poisson regression is modeling the mean leukemia response rate. Because the response rate in a person year is assumed to be linearly related to the cumulative ppm-years for that person year, the mean response rate in the highest dose group is proportional to the mean cumulative ppm-years in that dose group. Therefore, using the average ppm-year for the highest dose group is more appropriate than using an arbitrary value like 250 ppm-years. Although we contend it would be more mathematically correct to use the mean for all intervals, in this case using the mid-point instead of the average for the non-highest dose intervals of cumulative 1,3-butadiene exposure had very little effect on the estimates of the cancer potency.
2.2. Characterizing the excess cancer risk for a 70 -year lifetime In quantifying the potential excess cancer risks due to 1,3-butadiene exposure, EPA departed from their traditional default assumption of a 70-year lifetime and assumed an 85-year lifetime. According to the National Center for Health Statistics [9], a person born in the USA in 1990 is only expected to live 75.4 years (males 72.0 years and females 78.8 years). The quantitative impact of switching from the traditional
R.L. Sielken, Jr, C. Valdez-Flores / Chemico-Biological Interactions 135–136 (2001) 637–651 641
70-year lifetime to an assumed 85-year lifetime is to increase the risk characterization approximately threefold. Specifically, using EPA’s linear model for the rate ratio (1+ 0.0068× cumulative ppm-years), the added cancer risk associated with a continuous environmental exposure (24 h a day, every day of a lifetime) to 0.0001 ppm 1,3-butadiene is 0.88 ×10 − 6 (0.88 in one million) for 85 years at risk (age 0 to age 85) as opposed to only 0.32× 10 − 6 (0.32 in one million) for 70 years at risk (age 0 to age 70). Similarly, the added cancer risk is only 0.48× 10 − 6 (0.48 in one million) for 75 years at risk (age 0 to age 75). The risk for 75 years is 55% of the risk for 85 years. The risk for 70 years is only 36% of the risk for 85 years. By changing the standard lifetime from 70 years to 85 years, EPA has increased the risk characterization by 2.75-fold, as shown in Table 1.
2.3. Simplified explanation of the lifetime extra cancer risk calculation and its implications The following is a simplified explanation of the extra risk calculation. This explanation shows how the 0.0068 multiplier in the linear model for the rate ratio leads to a cancer potency of 0.0087 per ppm. The form of the calculation indicates why an increase from a 70-year lifetime to a longer lifetime results in a disproportionate increase in the lifetime extra unit cancer risk when the relevant dose metric is assumed to be cumulative ppm-years and it is assumed that all exposures before a specific age cause a multiplicative increase in the leukemia hazard rate for that age. In addition, the form of the calculation explains why the value of the lifetime excess cancer risk is substantially impacted by the background hazard rate. For simplicity, the adjustment for competing risks is omitted from the following explanation. The adjustment for competing risks (although somewhat complicated mathematically) was included (as it should have been) in the actual calculation of the cancer potency of 0.0087. Apart from the adjustment for competing risks, the lifetime extra cancer risk at c ppm is calculated as follows:
Table 1 Impact of the length of the lifetime on the estimation of the additional risks using EPA’s linear rate ratio Length of lifetime in years
Additional risk of leukemia at 0.0001 ppm
Percentage of additional risk for a lifetime of 85 years
70 75 80 85
0.3205×10−6 0.4826×10−6 0.6776×10−6 0.8811×10−6
36% 55% 77% 100%
(100%/2.75) (100%/1.83) (100%/1.30) (100%/1.00)
642 R.L. Sielken, Jr, C. Valdez-Flores / Chemico-Biological Interactions 135–136 (2001) 637–651
Extra Cancer Risk =Background Rate for the Specified Response in Year 1× 0.0068× 1 year × c ppm +Background Rate for the Specified Response in Year 2 × 0.0068 × 2 years× c ppm… + Background Rate for the Specified Response in Year 70× 0.0068 ×70 years× c ppm… +Background Rate for the Specified Response in Year 85× 0.0068 ×85 years× c ppm and, approximately, Extra Cancer Risk= c ppm × 0.0087 Because the cumulative dose increases with age for constant environmental exposures, the multiplier (here, 0.0068× age) of the background hazard rate increases as the age increases. Also, for many responses (including leukemia) the background hazard rate increases with age. Thus, the contribution to the lifetime extra cancer risk increases in magnitude as the age increases. An increase from a 70-year lifetime to an 85-year lifetime results in a disproportionate increase in the lifetime extra cancer risk. The form of the calculation also indicates how the value of the cancer potency is strongly determined by the choice of the Background Hazard Rate and the value that is substituted for ‘Background Rate for Specified Response in Year – – – ’ in the calculation of the cancer potency. For example, if this rate is halved, then the cancer potency is halved.
2.4. Characterizing the excess cancer risk for en6ironmental exposures based on the EPA Exposure Factors Handbook’s 6alue for the a6erage amount of air inhaled per day In converting from occupational exposure (8 h per day, 240 days per year) to continuous environmental exposure (24 h per day, 365 days per year), EPA assumed that humans inhale 20 m3 per day. This assumption is equivalent to assuming that environmentally exposed humans are approximately three times more exposed than occupationally exposed humans, because (20 m3/10 m3) × (365 days/240 days)= 3.042. Here, ‘three times more exposed’ means, for example, environmental exposure to 1 ppm results in three times more 1,3-butadiene intake than occupational exposure to 1 ppm. In other words, 1 ppm-year of environmental exposure is approximately equivalent to 3 ppm-years of occupational exposure.
R.L. Sielken, Jr, C. Valdez-Flores / Chemico-Biological Interactions 135–136 (2001) 637–651 643
Fig. 1. Distribution of the cumulative dose (ppm-years) of the workers that died with leukemia based on the updated and original estimates of 1,3-butadiene exposure.
Combining the average adult inhalation rates for men and women listed in the 1989 EPA Exposure Factors Handbook [6], the American Industrial Health Council Exposure Factors Sourcebook [7] recommends that the ‘‘daily inhalation rate for adults for use in risk assessments is 18 m3/day.’’ The 1997 EPA Exposure Factors Handbook [8] recommends even smaller values (11.3 m3/day for females and 15.2 m3/day for males). If EPA assumed that humans inhale 18 m3 per day, then environmentally exposed humans would be approximately 2.7 times more exposed than occupationally exposed humans, because (18 m3/10 m3) × (365 days/240 days)= 2.7375 Using an inhalation rate of 18 m3 instead of 20 m3 would have decreased the cancer potency by 10% from 0.0087 per ppm to 0.0078 per ppm (i.e. a 1.11-fold decrease). Using 1997 EPA values would result in larger decreases. 3. Updated UAB exposure database and analyses In 2000, Macaluso et al. [4] updated their exposure estimates for 1,3-butadiene, styrene, and dimethyldithiocarbamate among synthetic rubber workers. At the same time, Delzell et al. [10] reported a new assessment of lymphohematopoietic cancer and exposure to 1,3-butadiene and other chemicals. Under an agreement with the International Institute of Synthetic Rubber Producers, the UAB provided the authors with electronic files containing the updated epidemiological database of worker exposure histories. According to Macaluso et al.’s
644 R.L. Sielken, Jr, C. Valdez-Flores / Chemico-Biological Interactions 135–136 (2001) 637–651
final report [4], the updated exposure estimates are considerably higher than the original exposure estimates derived in 1995. Specifically, according to Macaluso et al., the updated job- and subject-specific exposures to 1,3-butadiene were up to one order of magnitude higher than the 1995 estimates. However, despite the significant difference between the magnitude of the updated 1,3-butadiene exposure estimates and the magnitude of the original exposures, the ranking of individual employees according to cumulative exposure levels using the original and updated estimates were equivalent. Thus, Macaluso et al. state that risk assessment analyses based on the updated exposure estimates are expected to decrease considerably when compared to risk characterizations based on the original exposure estimates. Fig. 1 shows the distribution of the cumulative dose (ppm-years) of the workers that died with leukemia based on the updated and the original estimates of 1,3-butadiene exposure. The average cumulative 1,3-butadiene exposure for workers that died with leukemia is 430 ppm-years based on the updated exposure estimates compared with an average of 66 ppm-years based on the original exposure estimates. The average ppm-years for leukemia decedents based on the updated exposure estimates is more than sixfold greater than the average ppm-years computed using the original exposure estimates.
4. Calculation of effective concentrations with updated UAB exposures and CMA-proposed changes The EPA used UAB’s 1995 linear Poisson regression model results to calculate the maximum likelihood estimate of the ‘effective concentration’ (EC01) and the 95% lower confidence limit on the effective concentration (LEC01) associated with an extra risk of leukemia of 0.01 (1%) from a lifetime exposure. Herein, the same form of the linear Poisson regression model is used with the new worker exposure estimates derived by Macaluso et al. [4] to update the EC01 and LEC01 values. Thus, the dose– response methodology is consistent with EPA’s draft risk assessment of 1,3-butadiene and the proposed cancer risk assessment guidelines of the EPA. The impacts of updated exposure values on the estimates of EC01 and LEC01 are indicated as well as the impacts of the three CMA-proposed changes discussed above. The dose – response component of the default linear Poisson regression model used by EPA and used herein is RR = 1 + i × dose where RR is the rate ratio (rate of leukemia deaths per person year with exposure equal to ‘dose’ divided by the rate of leukemia deaths per person year with zero exposure), i is the slope of the linear rate ratio function, and ‘dose’ is the cumulative exposure to 1,3-butadiene in ppm-years. In both EPA’s assessment and the analyses herein, the complete Poisson model incorporates the effect of possible covariates as follows:
R.L. Sielken, Jr, C. Valdez-Flores / Chemico-Biological Interactions 135–136 (2001) 637–651 645
Mean Leukemia Response Rate=Background Rate× Effect of Age × Effect of Calendar Year × Effect of Years Since Hire × Effect of Styrene × Linear Rate Ratio for Leukemia
(1)
where the background rate is the average leukemia deaths per person-year when the effect of the covariates is null (i.e. excluding the age effect, excluding the calendar year effect, excluding the styrene exposure, etc.). As in EPA’s assessment, the age covariate has values for each of the following intervals [40, 50), [50, 60), [60, 70), [70, 80), and 80 or more years of age; the calendar year covariate has intervals [1950, 1960), [1960, 1970), [1970, 1980), [1980, 1990), and [1990, 1991] where 1991 is the end of the follow-up; and the intervals for the year-since-hire covariate are [10, 20), [20, 30), and 30 or more years since hire. For styrene, the intervals are based on the quintiles used by Delzell et al. [10] in the final report (i.e. intervals for the cumulative exposure to styrene are defined such that each interval includes approximately the same number of deaths with leukemia). These intervals for styrene are 0, (0, 10.4], (10.4, 28.3], (28.3, 40.6], (40.6, 98.1], and more than 98.1 ppm-years. All the intervals for the covariates, except for the intervals for styrene exposure, are identical to those used in the Delzell et al. 1995 report [2] on which EPA based its risk assessment. The intervals to control for the effect of styrene exposures are constructed in the same way as in the 1995 report but are adjusted to reflect the updated exposure estimates reported by Macaluso et al. [4]. The cumulative 1,3-butadiene exposure in ppm-years is partitioned into intervals for the estimation of the i parameter in the linear rate ratio for leukemia. The intervals were defined so that each group of person years includes approximately the same number of leukemia deaths. The intervals are 0, (0, 35.6], (35.6, 69.3], (69.3, 180.5], (180.5, 265.6], (265.6, 400.9], (400.9, 791.9], and more than 791.9 ppm-years. Table 2 contains the cumulative 1,3-butadiene exposure intervals along with the observed number of leukemia deaths and person years included in each dose interval. In the evaluation of the linear rate ratio for leukemia, the average dose for each group is assumed to be equal to the mid-point of the interval (consistent with EPA’s draft risk assessment), except for the last group. The average dose for the last group is computed according to the CMA-proposed change and is equal to the average cumulative 1,3-butadiene exposure over all the person years with cumulative 1,3-butadiene greater than 791.9 ppm-years. In the maximum likelihood estimation for the Poisson regression model, the observation period for each of the 13 131 individuals in the analysis is subdivided into person years. The 234 430.6 person years are partitioned according to the age, calendar year, years since hire, cumulative styrene exposure, and cumulative 1,3-butadiene exposure intervals. (Cumulative exposures are computed from the beginning of exposure to the end of the person year.) There are 3600 partitions (five age intervals, five calendar year intervals, three years-since-hire intervals, six
Dose group (ppm-years)
0 (0, 35.6] (35.6, 69.3] (69.3, 180.5] (180.5, 265.6] (265.6, 400.9] (400.9, 791.9] More than 791.9 Summary
Point for parametric estimation
0.00 17.80 52.45 124.90 223.05 333.25 596.40 1776.71
Observed number of leukemia deaths
Number of person years in the group
Leukemia death rate per 10 000 person years Obs.
Pred.
7 8 7 7 7 8 7 8
47 911.2 59 206.4 27 923.1 43 414.2 15 655.1 14 648.9 14 961.3 10 710.4
1.46 1.35 2.51 1.61 4.47 5.46 4.68 7.47
1.46 1.47 1.90 2.69 3.19 3.82 4.41 8.59
59
234 430.6
2.52
2.52
646 R.L. Sielken, Jr, C. Valdez-Flores / Chemico-Biological Interactions 135–136 (2001) 637–651
Table 2 Summary of the cumulative 1,3-butadiene exposure (ppm-years) intervals used in the maximum likelihood estimation and the observed and predicted leukemia death rates
R.L. Sielken, Jr, C. Valdez-Flores / Chemico-Biological Interactions 135-136 (2001) 637–651 647
cumulative styrene exposure intervals, and eight cumulative 1,3-butadiene exposure intervals; 5× 5 ×3 ×6 × 8 =3600). The observed data for the maximum likelihood estimation are, for each of the 3600 partitions, the number of person years and the number of those years in which a leukemia death occurred. (Table 2 indicates a summary of the observed data (the sums of the values for the 450 partitions for each of the eight cumulative 1,3-butadiene exposure intervals) but does not indicate the values for each of the 3600 partitions individually and explicitly considered in the maximum likelihood estimation.) The likelihood is the product over the partitions of the Poisson probability of the observed number of leukemia deaths in the partition; namely, u xe − u/x! where x is the observed number of leukemia deaths in the partition and u is the mean response rate for the partition (i.e. the number of person years in the partition multiplied by the mean leukemia response rate (Eq. (1)) for a person year in the partition). The mean leukemia response rate (Eq. (1)) for a person year in a partition contains a parameter for the background rate, the age interval, the calendar-year interval, the years-since-hire interval, the styrene exposure interval, and the parameter in the linear rate ratio for leukemia. In the maximum likelihood estimation, 17 parameters are estimated; specifically, one parameter for the background rate, four parameters for the five age intervals, four parameters for the five calendar-year intervals, two parameters for the three years-since-hire intervals, five Table 3 Parameters in the Poisson regression model and their maximum likelihood estimates Multiplicative effect
Interval
Background rate
0.00006
Age (years)
[40, 50) [50, 60) [60, 70) [70, 80) 80 or more
Calendar year
[1950, [1960, [1970, [1980, [1990,
Years since hire
[10, 20) [20, 30) 30 or more
Styrene exposure (ppm-years)
Parameter value
1960) 1970) 1980) 1990) 1991]
0 (0, 10.4] (10.4, 28.3] (28.3, 40.6] (40.6, 98.1] More than 98.1 i: multiplier of cumulative 1,3-butadiene exposure in linear rate ratio for leukemia
1.0 1.57854 1.95228 5.67143 11.68332 1.0 0.33776 0.61959 0.60573 1.35311 1.0 2.72331 1.26667 1.0 0.91376 1.47260 3.07787 1.50670 1.54735 0.00157
648 R.L. Sielken, Jr, C. Valdez-Flores / Chemico-Biological Interactions 135–136 (2001) 637–651
Fig. 2. Comparison of the observed and predicted leukemia death rates per 10 000 person years.
parameters for the six styrene intervals, and one i parameter. (Because the model contains a background rate, the multiplicative effect of each covariate is, without loss of generality, normalized to 1.0 for the first interval and is an estimated parameter value for each remaining interval.) The maximum likelihood estimates of the parameters are in Table 3. A comparison of the observed and predicted leukemia response rates (summarized for each cumulative 1,3-butadiene exposure interval) is given in Table 2 and Fig. 2. The maximum likelihood estimate of the linear rate ratio using the Poisson regression model described above is RR =1 + 0.0016 × Cumulative 1,3-Butadiene Exposure (ppm-years) The maximum likelihood estimate of the rate ratio is used to estimate the lifetime excess risk of leukemia mortality for specified levels of continuous exposures to 1,3-butadiene. These lifetime risks are computed using the relative rate estimates and an actuarial approach that takes into account the 1990 U.S. background rate of leukemia deaths and the effects of competing causes of death reported in the 1990 U.S. population census [9]. Environmental exposures to 1,3-butadiene are assumed to be continuous during the entire lifetime. The lifetime of 70 years suggested as one of the CMA-proposed changes is used in this risk characterization. The occupational 1,3-butadiene exposures in the UAB epidemiology study are converted to continuous environmental exposures by multiplying the occupational exposure estimates by a factor that accounts for the volume of air inhaled per day (10 m3 in the workplace/18 m3 in the environment) and by another factor that accounts for the number of days exposed per year (240 days in the workplace/365 days in the environment).
R.L. Sielken, Jr, C. Valdez-Flores / Chemico-Biological Interactions 135–136 (2001) 637–651 649
The Poisson regression model estimate of the effective concentration EC01 satisfies 0.01 = [P(EC01) − P(0)]/[1 − P(0)] where P(EC01) is the probability of leukemia death by age 70 if continuously exposed to EC01 ppm of 1,3-butadiene. The actuarial approach for the computation of P(EC01) is based on P(EC01) =Sum over the years (t) from birth to 70 {[1990 U.S. Background Leukemia Death Rate at Age t] × [Rate Ratio Evaluated for Cumulative Dose at EC01] × [U.S. Probability of not Dying from a Competing Cause of Death Before Age t]}
The best fitting linear Poisson regression model implies that the maximum likelihood estimate of the effective concentration EC01 corresponding to an extra risk of 0.01 (1%) from a lifetime of continuous exposure to 1,3-butadiene is EC01 =15.1 ppm The equivalent cancer potency based on the default procedure of linearly extrapolating the extra risk of 0.01 at EC01 back to zero extra risk at dose zero is Unit Risk= 0.01/EC01 =0.01/15.1 =0.00066 per ppm The new EPA guidelines emphasize the use of effective concentrations as the points of departure for low-dose risk characterization. The effective concentration corresponding to an excess risk of 0.10 (10%) may be reasonable for animal studies, but it is not reasonable for human epidemiologic studies. Workers in epidemiologic studies usually experience exposure levels that are well below the EC10. Thus, for human epidemiologic studies, excess risks of 0.01 or below are more appropriate than 0.10. Furthermore, the estimated exposure levels corresponding to a 0.01 excess risk are more consistent and less variable than the estimated exposure levels associated with 0.10 excess risk with respect to alternative forms of the Poisson regression model (i.e. alternative forms of the relative ratio function of the cumulative exposure). EPA used the 0.01 excess risk level to compute the effective concentration in their draft risk assessment. The 95% lower confidence limit (LEC01) on the EC01 is computed using the 95% upper confidence limit on the slope parameter in the linear rate ratio. The 95% upper confidence limit on the slope is computed using EPA’s default procedure that is based on the likelihood ratio technique. The resulting LEC01 is equal to LEC01 =9.0 ppm with the corresponding upper bound on cancer potency equal to Upper Bound on Unit Risk= 0.01/9.0 = 0.0011 per ppm
650 R.L. Sielken, Jr, C. Valdez-Flores / Chemico-Biological Interactions 135–136 (2001) 637–651
The effective dose EC01 of 15.1 ppm includes the changes proposed to the SAB and EPA by the CMA. If the CMA-proposed changes had not been included, then the estimated EC01 would have been 2.8 ppm. Overall, the impact of using the three CMA-proposed changes is a 5.4-fold increase in the EC01 from 2.8 to 15.1 ppm and a 5.4-fold decrease in the maximum likelihood estimate of the unit risk from 0.0036 to 0.00066 per ppm. The first CMA-proposed change (namely, the use of the average cumulative exposure for the last dose interval) increases the EC01 by approximately 1.76-fold. This increase of 1.76-fold in the EC01 using the updated exposure estimates is greater than the increase of 1.30-fold when the original exposure estimates were used. The other two CMA-proposed changes (namely, the use of a daily inhalation rate of 18 m3 and a lifetime of 70 years) increase the EC01 by approximately 1.11and 2.74-fold, respectively, as before.
5. Comparison of new and previous cancer risk characterizations New leukemia risk characterizations have been developed using a methodology that is consistent with EPA’s draft risk assessment of 1,3-butadiene and EPA’s proposed cancer risk assessment guidelines. The new risk characterizations use the updated exposure estimates to 1,3-butadiene and incorporate the changes proposed by the CMA for the EPA’s draft risk assessment of 1,3-butadiene. The risk characterization published by EPA in their draft risk assessment of 1,3-butadiene estimated an EC01 equal to 1.16 ppm (or, equivalently, a cancer potency of 0.0087 per ppm). The risk characterization with updated exposure concentrations estimates an EC01 equal to 2.8 ppm (or a cancer potency of 0.0036 per ppm). That is, the incorporation of the updated exposure values in the estimation of linear Poisson regression model increases the EC01 (i.e. reduces the estimated cancer potency of 1,3-butadiene) by approximately a factor of 2.4. Incorporating the CMA-proposed changes increases the EC01 from 2.8 ppm to 15.1 ppm. This is an additional 5.4-fold increase over the estimated EC01 before the CMA-proposed changes are included. Overall, the EC01 (15.1 ppm) estimated in the present risk characterization is approximately 13-fold greater than the EC01 (1.16 ppm) estimated in the risk characterization presented by EPA in their draft risk assessment of 1,3-butadiene. The potency value of 0.0087 per ppm estimated with the original exposure data is more than 13-fold greater than the potency value of 0.00066 per ppm estimated by us using the updated, more reliable, exposure estimates and the CMA-proposed changes.
6. Conclusions and future research In 1998 EPA’s draft risk assessment for 1,3-butadiene estimated an EC01 equal to 1.16 ppm and a cancer potency of 0.0087 per ppm based on UAB’s human epidemiologic data. Using EPA’s methodology, UAB’s updated exposure estimates,
R.L. Sielken, Jr, C. Valdez-Flores / Chemico-Biological Interactions 135–136 (2001) 637–651 651
and the three CM-proposed changes, the estimated EC01 is 15.1 ppm, and the cancer potency is 0.00066 per ppm. UAB’s updated exposure estimates, which have been extensively reviewed, are a significant step in the direction of more reliable estimation. The reliability of risk characterizations of 1,3-butadiene using epidemiologic data, however, can still be improved. Currently, the cohort includes exposures and follows workers through the end of 1991. The cohort can be followed for a longer period to include more accurate estimation in the older years of life. Although peak 1,3-butadiene exposures and average 1,3-butadiene exposures for concentrations above and below 100 ppm have been estimated, the role of these exposures in lymphohematopoietic cancers and cancer dose–response modeling needs to be clarified. The most biologically relevant dose metric (e.g. combination of the duration and intensity of exposure, toxicokinetic and toxicodynamic doses) for dose–response modeling and risk characterization needs to be determined. Similarly, the effect of a lag in exposures and the effect of latency need to be further investigated. References [1] U.S. Environmental Protection Agency, Health Risk Assessment of 1,3-Butadiene — External Review Draft, National Center for Environmental Assessment, Office of Research and Development, 63 Fed. Reg. 7167 (February 12, 1998) Publication NCEA-W-0267, Washington, 1998. [2] E. Delzell, N. Sathiakumar, M. Macaluso, M. Hovinga, R. Larson, F. Barbone, C. Beall, P. Cole, J. Julian, D.C.F. Muir, A follow-up study of synthetic rubber workers: A report submitted to the International Institute of Synthetic Rubber Producers, 1995. [3] E. Delzell, N. Sathiakumar, M. Hovinga, M. Macaluso, J. Julian, R. Larson, P. Cole, D. Muir, A follow-up study of synthetic rubber workers, Toxicology 113 (1996) 182 – 189. [4] M. Macaluso, R. Larson, J. Lynch, S. Lipton, E. Delzell, Historical estimation of exposure to 1,3-butadiene, styrene and dimethyldithiocarbamate among synthetic rubber workers: Final report submitted to the International Institute of Synthetic Rubber Producers, 2000. [5] Chemical Manufacturers Association, Comments of the olefins panel of the Chemical Manufacturers Association on EPA’s Draft Health Risk Assessment for 1,3-Butadiene, 1998. [6] U.S. Environmental Protection Agency, Office of Health and Environmental Assessment, Exposure Factors Handbook, EPA/600/08-89/043, Washington, 1989. [7] American Industrial Health Council (AIHC), Exposure Factors Sourcebook, AIHC, Washington, 1994. [8] U.S. Environmental Protection Agency, Office of Research and Development, Exposure factors handbook, Volume I — General Factors, EPA/600/P-95/002Fa, Washington, 1997. [9] U.S. Department of Health and Human Services. Vital Statistics of the United States 1990, Volume II — Mortality, Part A. [10] E. Delzell, M. Macaluso, N. Sathiakumar, R. Matthews, Lymphohematopoietic cancer among workers exposed to 1,3-butadiene, styrene and dimethyldithiocarbamate in the synthetic rubber industry, Final report submitted to the International Institute of Synthetic Rubber Producers, 2000.
.