Estimation of the cumulated exposure to polychlorinated dibenzo-p-dioxins/furans and standardized mortality ratio analysis of cancer mortality by dose in an occupationally exposed cohort.

For a cohort of 1189 male German former herbicide and insecticide workers with exposure to polychlorinated dibenzo-p-dioxins and -furans (PCDD/F), we report an extended standardized mortality ratio (SMR) analysis based on a new quantitative exposure index. This index characterizes the cumulative lifetime exposure by integrating the estimated concentration of PCDD/F at every point in time (area under the curve). Production department-specific dose rates were derived from blood levels and working histories of 275 workers by applying a first-order kinetic model. These dose rates were used to estimate exposure levels for all cohort members. Total mortality was elevated in the cohort; 413 deaths yielded an SMR of 1.15 (95% confidence interval [Cl] 1.05, 1.27) compared to the mortality of the population of Germany. Overall cancer mortality (n = 124) was significantly increased (SMR = 1.41, 95% Cl 1.17, 1.68). Various cancer sites showed significantly increased SMRs. The exposure index was used for an SMR analysis of total cancer mortality by dose. For 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) a significant trend (p = 0.01) for the SMRs with increasing cumulative PCDD/F exposure was observed. The SMR in the first exposure quartile (0-125.2 ng/kg x years) was 1.24 (95% Cl 0.82, 1.79), increasing to 1.73 (95% Cl 1.21, 2.40) in the last quartile (> or = 2503.0 ng/kg x years). For all congeners combined as toxic equivalencies (TEQ) using international toxic equivalency factors, a significant increase in cancer mortality was observed in the second quartile (360.9-1614.4 ng/kg x years, SMR 1.64; 95% Cl 1.13, 2.29) and the fourth quartile (> or = 5217.7 ng/kg x years TEQ, SMR 1.64, 95% Cl 1.13, 2.29). The trend test was not significant. The results justify the use of this cohort for a quantitative risk assessment for TCDD and to a lesser extent for TEQ.

Several attempts have been made to estimate the magnitude of the cancer risk at environmental levels (11) or to determine a safe dose, i.e., that at which no adverse effect is expected (12,13). Available risk estimates were derived mainly from animal carcinogenicity data, especially the Kociba et al. (14) study. Results of human studies are not yet being used because of a lack of dose-response data on which to quantify the magnitude of risk at certain doses.
A first approach to quantitative analysis of a dose-response relationship on cancer was published recently (15) for a cohort of polychlorinated dibenzo-p-dioxins/furans (PCDD/F)-exposed workers (4). An estimate for the PCDD/F blood levels at the end of employment was derived for the whole cohort using blood-level data for a subgroup of 190 workers. From these blood levels production department-specific linear average yearly increases in PCDD/F blood concentrations (nanogram/kilogram/year) were estimated by regressing the blood levels at the end of exposure (backcalculated from the measured levels using a first order kinetic assumption) on the duration of work in different production departments. These estimates were then used to calculate the expected PCDD/F blood levels at the end of exposure for all cohort members by multiplying the working times by the estimates of the yearly increase. Using these dose parameters a dose-response relationship for cancer and exposure to PCDD/F was demonstrated.
The use of these results to assess cancer risk at background levels suffers from several restrictions. First, the estimation procedure used in this former paper (15) did not take into account the elimination during exposure, which results in a nonlinear increase ( Figure 1). Second, the course of the working times in different departments was not considered in the statistical model. This raises the possibility that a worker could have worked in a high-contamination area first, followed by working a long time in a department with low exposure. This would assign the worker a low blood level To illustrate the above situation, Fi 1 shows the course of the TCDD b level for a hypothetical person with a rate of 1 ng/kg/year up to 20 years o0 followed by a 10-year exposure to a rate of 20 ng/kg/year measured 20 y after the end of exposure compared tC course of the TCDD blood level for a son with a dose rate of 1 ng/kg/year u 50 years of age. Several dose pararn choices are available: the maximum centration a person experienced over t the level at the end of exposure (not no sarily identical with the maximum con tration over time), or the integral ol concentration over time (area under curve). The area under the curve was sen for risk assessment analysis becau considers variations in the concentra over time and reflects cumulative life exposures to dioxins and furans (16). This paper presents basic considerai using the cohort for risk estimation describes construction of the dose var and a dose-response analysis relation for cancer using the mortality of the p lation of Germany as a reference. A det; dose-response analysis taking into acc several covariates and coexposure, espec to beta-hexachlorocyclohexane (,B-H( and an estimation of cancer risk at envi mental background levels are presente Becher et al. (17).

Materials and Methods
The basic methods followed in this si follow Manz et al. (4) and Flesch-Janj End ofexposure. at the nearest time point for which data were available (22)(23)(24) were induded in the estimation ofdose rates. First, measured concentrations of each congener diminished by the median of the background concentration were backcalculated to the end of employment under the assumption of a first order ime of mansuramnnt elimination process. A first order kinetic equation was developed linking blood levels and working histories to produce production department-specific dose rates for every congener. Dose rates were used to estimate the concentration of every congener at every a. z 30 *5-Z^50 point in time for all cohort members. The cumulated PCDD/F levels expressed as Age, years nanogram/kilogram blood fat x years were calculated by integration and used in the els for two persons with different exposures.
standardized mortality ratio (SMR) analysis. Details of this modeling procedure are tsure. al. (15). The cohort is composed of 1189 presented in "Appendix." t the males employed on 1 January 1952 or later o be for at least 3 months. Follow-up ended 31 culatlon ofStdaIzed oxins December 1992.
Mortality Ratios Standardized mortality ratios were igure EimationofDoseRates calculated using the gender-, age-, and blood Measurements of PCDD/F levels in blood calendar year-specific mortality rates of dose (n= 320) or adipose tissue (n= 62) were the German population for 1952 to 1992 f age available for 275 workers (39 females, 236 (Federal Statistical Agency, Germany, perdose males). Two or three measurements were sonal communication). Confidence intervals years available for some workers, yielding a total (CIs) were calculated assuming the Poisson the of 382 blood samples. The blood levels were distribution for the observed cases. SMRs per-determined by the ERGO laboratory were estimated for the total cohort and for p to (Hamburg, Germany). Measurement methexposure levels categorized into four groups ieter ods used are described in Stephens et al. according to the quartiles of the calculated con- (18) and the adipose tissue concentrations area under the curve above background at time, are described in Beck et al. (19). The the end of the follow-up. Person-years were ecesconcentrations are reported in nanogram/ calculated taking into account the course of icen-kilogram blood or adipose fat. Toxic equiva-the person through the exposure classes. f the lencies (TEQs) were calculated using the Trend tests were performed by linear regresr the international toxicity equivalency factors (Ision of the SMRs on the geometric means of cho-TEFs) (20). Working histories covering the the exposure groups weighted by the number se it duration of employment in 22 different ofobserved cases (25).

PCDDIF Blood Lls
Only workers whose concentration at the time of measurement exceeded the Bloodor adipose tissue-level data were 95% percentile of the German population available for 275 workers. Table 1 shows Abbreviations: Max, maximum; Min, minimum; SD, standard deviation; TEQM, international toxic equivalencies with TCDD; TEQO, international toxic equivalencies without TCDD; TSEXIT, time since end of employment (years). 'f more than one measurement was available the first was included in the calculation. the descriptive parameters for the PCDD/F entered the plant later (median 1967 vs and clean-up workers. These categories levels. The arithmetic mean for TCDD 1959), and left the plant later (median were combined together from the main was 101.3 ng/kg (minimum, 2.0; maxi-1983 vs 1968). departments in the subsequent analysis. mum, 2252 ng/kg). For the higher chlori-Workers with blood levels had longer . of Dose Rates nated PCDD/F without TCDD calculated periods of employment than the workers as international toxic equivalency (I-TEQ) without (median 9.2 vs 3 years). Table 2 also Estimated dose rates for TCDD are shown the mean was 89.3 (minimum, 5.0; maxi-shows the distribution ofworkers across pro-in Table 3. The highest dose rate was mum, 1131.9). Table 2 characterizes the duction departments. In general an adequate obtained for the trichlorophenol department group with available blood levels compared number of workers with blood levels were before the change in production process in to the group of workers without blood available, but this was not always true-1957 (3376.4 ng/kg blood fat/year). For levels. The first group is slightly younger, especially with regard to administration the 2,4,5-trichlorophenoxyacetic acid Table 2. Number of workers with and without data available on blood levels, age, and year at entry, of end, and duration of employment in every production department.  the data for different time periods were sparse. No gender-specific effect or effects of short durations of employment on dose rates were seen. Data for these latter analyses were also sparse. Figure 2, which illustrates the fit of the model, compares TCDD concentrations at the end of employment calculated from the measured levels (using "Appendix" equation 3) with TCDD concentrations predicted by the model. The symbols indicate different lengths of employment. The Spearman rank correlation coefficient for comparison of the estimated with the measured levels was 0.53.
Estimated dose rates for the higher chlorinated congeners without TCDD expressed as I-TEQs are shown in Table 3 (26). However, the blood levels of these congeners generally were within the range of the ubiquitous background levels and did not contribute substantially to the total TEQs. The Spearman rank correlation coefficients for the higher chlorinated congeners varied between 0.26 for heptadioxin and 0.60 for 1    Estimates for the other departments ranged between 0 for administration and 17.7 pg/liter blood/year for the lindane department. The Spearman rank correlation coefficient was 0.72 (p< 0.01). Figure  3 shows the estimated time/concentration curve for one of the cohort members.  1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 Year Figure 3. Estimated concentration over time for TCDD and TEG for one cohort member.

Standardized Mortlity Ratio Analysis
2.78). Estimates for several localizations were below 1; however, with the exception of colon cancer, these generally were not localizations with large numerical impacts on total cancer mortality. The SMR for cardiovascular diseases was slightly nonsignificandy elevated (1.06; 95% CI 0.90, 1.24). No increase was observed for nonmalignant respiratory diseases or for digestive diseases. The SMR for unnatural causes (accidents and suicides) was 1.79 (95% CI 1.35, 2.33). There were 24 ill defined or unknown causes of death, yielding an SMR of 2.59 (95% CI 1.66, 3.85). Table 5 shows the results of the SMR analysis using the estimated integrated TCDD concentration until the end of follow-up as dose parameter. A U-shaped relation between dose and mortality was observed for all causes of death. The linear trend test was not significant. The SMR for all cancer combined was 1.24 (95% CI 0.82, 1.79) for the first quartile (up to 125.2 ng/kg blood fat xyear) and increased from 1.34 in the second and third to 1.73 (95% CI 1.21, 2.40) in the fourth quartile (more than 2503.0 ng/kgxyear). The linear trend test was significant (p= 0.0 13). No trend was observed for lung cancer or for all hematopoietic and lymphatic cancers combined. Total TEQs also showed a U-shaped relation to total mortality ( Table 6). For total cancer the SMR in the first quartile (up to 360.9 ng/kgxyear) was 1.07 (95% CI 0.69, 1.58). A significant increase was observed in the second quartile (1.64; 95% CI 1.13, 2.29), which ranged up to 1614.4 ng/kgxyears. The Table 4. Standardized mortality ratios for selected causes of death using the mortality rates of the population of Germany as reference.

ICD-9
Cause  Discussion A significant 40% increase in total cancer mortality was observed for the whole cohort compared to that for the German population. Thus, cancer mortality increased in the 3 additional years of follow-up (1989 to 1992) from 1.24 to 1.41. This elevation in total cancer mortality was not restricted to one localization. Significant increases were observed for lung, all respiratory cancers, rectum, and hematopoietic and lymphatic cancers, especially lymphosarcomas, which belong to the non-Hodgkin's type of lymphomas. No case of soft-tissue sarcoma was observed, although 0.33 were expected. Available blood levels revealed the cohort had substantial exposure to TCDD. Mean concentration at the time of measurement was 101.3 ng/kg TCDD, with a maximum level of 2252 ng/kg. There was also substantial exposure to higher chlorinated congeners, with a mean of 89.3 ng/kg TEQ (without TCDD) and a maximum of 1131.9 ng/kg. Estimated dose rates derived from these measurements allowed estimation of the maximal concentration for each worker during his or her period of observation. For TCDD a mean of 340.5 ng/kg was observed. Highest concentrations were estimated for three workers with values between 10,000 and 13,000 ng/kg. The mean of the estimated maximum concentrations for total I-TEQ was 473.5 ng/kg, with a maximum value of 13,179 ng/kg I-TEQ.
With the exception of ,-HCH [see Becher et al. (17)], other potential confounders like smoking habits and exposure to other carcinogenic or suspected carcinogenic substances could not be addressed directly. With regard to smoking, we showed that blood-level estimates were not correlated with smoking status for a subgroup of workers (15) for whom these data were available. In addition, calculating total cancer mortality without lung cancer cases for the TCDD exposure groups yielded an even more pronounced trend (SMRs for TCDD quartiles I  (15)].
An important question in dose-response analysis is whether exposure measurements reflect the exposure of all cohort members with sufficient accuracy. First, the relationships between the production departmentspecific estimates are in good agreement with the expectation from the chemistry of the production processes and the available data on the contamination of products, buildings, and waste. The highest dose rates for TCDD were obtained for the 2,4,5-T and 2,4,5-TCP departments, as expected.
Measurements indicate contamination of 2,4,5-T acid in the parts per million range in former years, whereas in waste from 2,4,5-T production concentrations up to 3360 pg/kg TCDD were detected. In contrast, the octachlorodibenzodioxin concentrations of up to 7200 pg/kg were comparably low. Conversely, the highest dose rates for the higher chlorinated congeners were obtained for the thermic decomposition department, where measurements of smelting showed hexaand octadioxin concentrations up to 32x 106 pg/kg but only 500 pg/kg TCDD (21). For this department the highest dose rates for the higher chlorinated congeners were detected.
For some departments we observed positive but nonsignificant dose rates. We were aware of the possibility of overfitting the model; however, the estimates were all in good agreement with a priori knowledge of exposure levels so we decided to use them. No information was available on job duties within the departments or on potential accidental exposure. These factors could yield some additional variations.
Another critical point is the assumption of a first order elimination kinetic. Available human data (26,27) indicate that this assumption is reasonable, at least in the observed dose range. No human data were available on the form of absorption kinetics that could have been considered in the exposure model. There were few workers with short durations of employment and comparably high levels of PCDD/F. Whether this could be attributed to individual exposure histories, i.e., accidental or time/concentration-dependent modulation of PCDD/F absorption, remains unclear. SMR analysis revealed a significant trend for total cancer mortality with increasing estimated cumulated TCDD levels. This result is supported by a study from the Netherlands (28) and the latest results on cancer mortality within an accidental German cohort (5). The dose estimates in the Netherlands study indirectly support the dose-rate estimation presented in this paper in that the Netherlands results were roughly in the same order of magnitude, though a different estimation technique was used. For total TEQ the trend was not as pronounced as for TCDD.
In summary, we observed an elevated risk of total cancer mortality in a cohort with high exposure to PCDD/F. A dosedependent effect was observed for estimated TCDD levels on total cancer mortality by SMR analysis using an index that characterizes cumulated exposure to TCDD and the higher chlorinated congeners. We conclude that use of these data for quantitative cancer risk assessment is justified.