Differences in baseline lung cancer mortality between ...

Differences in baseline lung cancer mortality between the German uranium miners cohort and the population of the former German Democratic Republic (1960–2003)

Radiation and Environmental Biophysics ISSN 0301-634X Volume 50 Number 1 Radiat Environ Biophys (2010) 50:57-66 DOI 10.1007/s00411-010-0332y

1 23

Your article is protected by copyright and all rights are held exclusively by SpringerVerlag. This e-offprint is for personal use only and shall not be self-archived in electronic repositories. If you wish to self-archive your work, please use the accepted author’s version for posting to your own website or your institution’s repository. You may further deposit the accepted author’s version on a funder’s repository at a funder’s request, provided it is not made publicly available until 12 months after publication.

1 23

Author's personal copy Radiat Environ Biophys (2011) 50:57–66 DOI 10.1007/s00411-010-0332-y

ORIGINAL PAPER

Differences in baseline lung cancer mortality between the German uranium miners cohort and the population of the former German Democratic Republic (1960–2003) Linda Walsh • Florian Dufey • Matthias Mo¨hner Maria Schnelzer • Annemarie Tschense • Michaela Kreuzer

•

Received: 31 March 2010 / Accepted: 7 September 2010 / Published online: 25 September 2010 Springer-Verlag 2010

Abstract A previous analysis of the radon-related lung cancer mortality risk, in the German uranium miners cohort, using Poisson modeling techniques, noted internal (spontaneous) rates that were higher on average than the external rates by 16.5% (95% CI: 9%; 24%). The main purpose of the present paper is to investigate the nature of, and possible reasons for, this difference by comparing patterns in spontaneous lung cancer mortality rates in a cohort of male miners involved in uranium extraction at the former Wismut mining company in East Germany with national male rates from the former German Democratic Republic. The analysis is based on miner data for 3,001 lung cancer deaths, 1.76 million person-years for the period 1960–2003, and national rates covering the same calendar-year range. Simple ‘‘age–period–cohort’’ graphical analyses were applied to assess the main qualitative differences between the national and cohort baseline lung cancer rates. Some differences were found to occur mainly at higher attained ages above 70 years. Although many occupational risk factors may have contributed to these observed age differences, only the effects of smoking have been assessed here by applying the Peto–Lopez indirect method for calculating smoking attributability. It is inferred that the observed age differences could be due to the greater prevalence of smoking and more mature smoking

L. Walsh (&) F. Dufey M. Schnelzer A. Tschense M. Kreuzer Federal Office for Radiation Protection, Department ‘‘Radiation Protection and Health’’, Ingolsta¨dter Landstr. 1, 85764 Oberschleissheim, Germany e-mail: [email protected] M. Mo¨hner Federal Institute for Occupational Safety and Health (BAuA), No¨ldnerstraße 40–42, 10317 Berlin, Germany

epidemic in the Wismut cohort compared to the general population of the former German Democratic Republic. In view of these observed differences between external population-based rates and internal (spontaneous) cohort baseline lung cancer rates, it is strongly recommended to apply only the internal rates in future analyses of uranium miner cohorts.

Introduction The characteristics of the German ‘‘Wismut’’ uranium miners cohort have already been described elsewhere (Kreuzer et al. 2009a). It is currently the largest miners cohort study with 3,016 deaths occurring from lung cancer and almost 2 million person-years of observation for the full follow-up period from 1.1.1946 to 31.12.2003. Several analyses on this cohort, based on detrimental health effects’ data pertaining to 58,987 male former employees, have recently been published (Kreuzer et al. 2008, 2009b; Schnelzer et al. 2010; Walsh et al. 2010a, b). There are several occupational risk factors for detrimental health effects which are relevant to the cohort members, including exposure to radon, gamma radiation, long-lived radionuclides (Lehmann et al. 1998), fine dust, arsenic dust and quartz fine dust (Dahmann et al. 2008), diesel and asbestos (Bru¨ske-Hohlfeld et al. 2006). In radiation epidemiological cohorts assessed by cancer excess relative (or absolute) risk models, the baseline rates are defined as the cancer rates that would have occurred in the absence of radiation, i.e., spontaneously. The Wismut study includes a large number of non-radiation-exposed cohort members, contributing 25% and 21% of the total person-years for the full cohort and the sub-cohort from 1960, respectively. This substantial non-exposed

123

Author's personal copy 58

proportion forms a good contribution to the internal comparison group for a reliable determination of baseline cancer rates that can be compared with national rates for the former German Democratic Republic (GDR) that are available from 1960. Note, however, that the internally estimated baseline rates also contain a (statistical) contribution from the exposed cohort members. To illustrate this point further, 8% of the total lung cancers occur in the unexposed group, but a previous Poisson modeling analysis (Walsh et al. 2010a) found that 53% of the total lung cancers contributed to the baseline. This previous analysis of the lung cancer risk in German uranium miners, from exposure to radon, noted internal (spontaneous) rates that were higher on average than the external rates by 16.5% (95% confidence interval (CI): 9%; 24%) (Walsh et al. 2010a). The aim of the present analyses is to investigate the nature and possible causes of this result by comparing spontaneous lung cancer mortality rates in cohort members, i.e., miners involved in uranium extraction at the former Wismut mining company in East Germany, with male national rates for the former GDR. Due to the limited availability of rates for the former GDR mentioned above, the analysis here is based on the subset of Wismut miner cohort data covering the period 1960–2003 and national rates for the same period. Simple ‘‘age–period– cohort’’ graphical analyses, which involve the mortality rates per 100,000 persons by age at death, period of death and birth cohort, are applied to assess the main qualitative differences between the national and cohort rates. An indirect procedure (the ‘‘Peto–Lopez method’’) for calculating the proportion of lung cancer deaths attributable to smoking via a smoking impact factor (SIF) (Peto et al. 1992, 1994 and Powles 2000) has also been applied. Methods are also applied, in the quantitative analyses involving standardized mortality ratios (SMR) and excess relative risk (ERR) models, to adjust the cohort rates for unknown causes of death because these rates show some systematic variation with calendar time, age at death and radon exposure class.

Materials and methods Cohort definition, time periods and mortality follow-up Full details of the cohort have already been given (Kreuzer et al. 2009b; Walsh et al. 2010a). Every cohort member contributes to the number of person-years starting 180 days after the date of first employment and ending at the earliest of date of loss to follow-up, date of death or end of followup (31.12.2003). In the present analyses, the former disease codes of the comparison of external baseline rates for the

123

Radiat Environ Biophys (2011) 50:57–66

GDR were recoded via earlier ICD revisions to the 10th ICD code (WHO 1992), which was applied throughout. This recoding process was complicated by several revisions to ICD codes during the period of data coverage and German reunification. Population lung cancer rates are not available just for the relevant mining region of Turingia and Saxony. Consequently, the external rates applied here cover the total area of the former GDR (including East Berlin) during the time period 1960–1997; in contrast, from 1998, the rates pertain to the former GDR states and the whole of Berlin. The codes used here in the various time periods are as follows: 1960–1967: GDR code number 735 for trachea, bronchus and primary lung cancers; 1968–1979: ICD 8, code number 162; 1980–1997: ICD 9, code number 162; 1998–2003: ICD 10, code number C33, C34 all for trachea, bronchus and lung cancers. Analysis Three methods were applied that all require the tabulation of the individual data as described below and in previous analyses (Walsh et al. 2010a, b). Simple ‘‘age–period– cohort’’ graphical analyses, which involve the mortality rates per 100,000 persons by age at death, period of death and birth cohort, are applied to assess the main qualitative differences between the national rates and the baseline cohort rates (i.e., the spontaneous lung cancer mortality rates after the radon-related lung cancer rates have been accounted for). In this type of graphical analysis, it is necessary to match not only the calendar period of 1960–2003 but also the birth cohorts, i.e., the birth cohorts from 1875 to 1900 need to be excluded from the external rates because the miner cohort includes only the birth cohorts from 1900. Two quantitative risk evaluation methods based on a known association between the lung cancer mortality risk and cumulative radon exposure have also been applied here. These are the simple SMR model (considered both with and without an exposure response) and a previously published continuous Poisson regression ERR model (Walsh et al. 2010b) that contains all of the substantial age- and time-related radon exposure riskmodifying factors, in which a 5-year lag was used in calculating the cumulative exposure to radon. This ERR model for lung cancer is our preferred main model from many models assessed by model selection techniques and is linear in radon exposure with exponential effect modifiers that depend on age at median exposure, time since median exposure and exposure rate (with only minimal confounding e.g. by dust and arsenic). Since the internally estimated background rates depend on the applied models, the simple SMR model served as a first indicator of differences between external and internal baseline rates and included overall and specific corrections

Author's personal copy Radiat Environ Biophys (2011) 50:57–66

59

for missing causes of deaths. The preferred ERR model was then applied in the further detailed investigations into differences in internal to external baseline rates, by extracting the baseline rates from the strata in this model for a direct comparison, exterior to the modeling procedure, with the actual external rates. The preferred ERR model did not include a correction for missing causes of death.

k*(a, y) denotes the external rates as a function of age and calendar year and k(a, y) denotes the observed rates in the miners cohort, then the SMR model can be written as

Data tabulations

RRða; y; wÞ ¼ b1 k ða; yÞ ð1 þ b2 ðwÞÞ

Tabulations of person-years at risk and cancer deaths were created with the DATAB module of the EPICURE software (Preston et al. 1993). Age at median exposure and time since median exposure were calculated with reference to median exposures, i.e., when half of the exposure cumulated up to a given date was reached. Cross-classifications were made by attained age, a, in 16 categories (\15, 15–\20, 20–\25, …, 85? years), individual calendar year, y, in 58 categories, age at median exposure, e, in seven categories (\20, 20–\25, 25–\30, 30–\35, 35–\40, 40–\45, 45? years), time since median exposure, t, in six categories (\5, 5–\10, 10–\15, 15–\20, 20–\25, 25? years) and cumulative radon exposure, w, in nine categories (0,[0–\10, 10–\50, 50–\100, 100–\200, 200–\500, 500–\1,000, 1,000–\1,500, 1500? WLM1). The exposure rate, er, calculated as the re-computed total cumulative working level months (WLM) divided by total duration (with a lag) at each attained age, on the assumption of 11 working months per year, was also categorized into six groups (0–\0.5, 0.5–\1, 1–\2, 2–\4, 4–\10, 10? WL).

to estimate the radon exposure (w) effect, based on the GDR external rates, assuming that the SMR for the background rates is identically equal to 1, i.e., b1 is fixed to unity during the optimization. In this case, b2 is a fit parameter that then gives the simple ERR per unit of radon exposure relative to the external GDR rates. It is also possible to test whether the external GDR rates are different from the internal baseline rates in the miners cohort by simply freeing the parameter b1 and repeating the optimization. The influence of missing causes of death has been investigated by fitting the models in Eqs. 1 and 2 with either no correction for missing causes of death or an average correction using one value of the proportion of causes of death that are known (Rittgen and Becker 2000) or a fine correction using the proportion of causes of death that are known either in each calendar-year period or in each Poisson cell. The Poisson cells here refer to the data elements obtained from tabulating the individual data into the groups described above. These methods assume that the missing lung cancers are evenly distributed throughout the covariable ranges in a way that is proportionate to the percentage of missing causes of death.

External rates and standardized mortality rates Mortality rates observed in the cohort were compared with those of the general male population in Eastern Germany, formerly the GDR. Since external rates were only available from 1960 onwards, the SMR analysis was limited to the follow-up period 1960–2003. Therefore, 15 lung cancers deaths with the corresponding person-years prior to 1960 were excluded. The first stage of the SMR analysis for lung cancer has been done here in the same way as described previously for extra-pulmonary cancers (Kreuzer et al. 2009b) with some extensions that allow a comparison of internal (miner cohort) and external (former GDR) baseline (spontaneous) rates. Two finer methods for accounting for the unknown causes of death in the miners cohort were also used. The simplest SMR model relates the rates in the population of interest (the miners cohort) to a multiple of the rates from the external population (the former GDR). If 1

One WLM of cumulative exposure corresponds to exposure to 1 WL during one month (170 h) and is equivalent to 3.5 mJh/m3.

kða; yÞ ¼ b k ða; yÞ

ð1Þ

where the b is a fit parameter and represents the SMR. An SMR [ 1 for the miners cohort is a known result (Walsh et al. 2010a). Therefore, it is instructive to fit a relative risk (RR) model ð2Þ

ERR parametric risk models (internal comparison group) The tabulated data were fitted to the following model (Walsh et al. 2010b)—if r(a, y, w, er, e, t) is the exposurespecific lung cancer mortality rate that depends on age, year, exposure, exposure rate, age at median exposure and time since median, and r0(a, y) = r(a, y, 0, 0, 0, 0) is the baseline disease rate for non-exposed individuals (w = 0, er = 0), then r ða; y; w; er; e; tÞ ¼ r0 ða; yÞ f1 þ ERRðw; er; e; tÞg

ð3Þ

where ERR is the excess relative risk factorized into a function of exposure and a modifying function: ERRðw; er; e; tÞ ¼ bw w exp½ww ðer 2:7Þ þ aðe 33Þ þ eðt 11Þ

ð4Þ

where and bw, ww, a and e are fit parameters, w is the cumulated radon exposure and er is the exposure rate;

123


e and t are calendar time–dependent variables, calculated with reference to the median time-lagged exposures. The model-centering constants, at an age at median exposure of 33 years and time since median exposure of 11 years, were chosen to match the mean Wismut cohort values—the choice of centering constants only serves to change the risk by a factor and has no influence on the goodness of fit of a particular model. Maximum likelihood with the AMFIT module of the EPICURE software (Preston et al. 1993) was used for the estimation of the fit parameters and the internal baseline rates in 928 strata of age attained (16 categories) and individual calendar year (58 categories), chosen to match the grouping that was available in the external rates. Out of these 928 possible strata, 708 actually contained rate data that were extracted for direct comparison with the national rates for the former GDR. Peto–Lopez indirect method for calculating deaths attributed to smoking Since data on smoking habits are not available for the majority of Wismut cohort members, it is useful to have a method for estimating the effects of smoking directly from the baseline lung cancer mortality rates. As suggested by Sir Richard Peto FRS and colleagues (Peto et al. 1992, 1994 and Powles 2000), the prevalence of smoking can be estimated indirectly by comparing the lung cancer mortality rates of interest with the lung cancer rates among current smokers and never smokers in a large prospective cohort study conducted by the American Cancer Society (ACS) CPS-II study from the 1980s (see Thun et al. 2008, open access article for direct links to the data in EXCEL format and relevant references). The two main assumptions here are that: the CPS-II lung cancer mortality rates for current smoker and never smoker are a valid approximation of the (unobserved) smoking-specific lung cancer mortality rates pertaining to the two sets of rates of interest here (the Wismut baseline rates and the external lung cancer rates of the former GDR); and current lung cancer mortality provides a better measure of the effect of lifetime tobacco smoking than smoking prevalence. Estimates of the fractions of tobacco-attributable deaths related to both the internal Wismut and external former GDR baseline rates were obtained by this indirect method. This was achieved by comparing both internal and external baseline lung cancer mortality rates with rates observed for current smokers or never smokers in the above-mentioned CPS-II cohort of men of European descent and deriving a ‘smoking impact factor’ (SIF). The SIF was calculated according to the following formula that pertains to economically developed countries:

123


SIF ¼ Lung cancer death rate in excess of never smokers in group of interest=Excess lung cancer death rate for known reference group of current smokers ¼ ðCLC NLC Þ= SLC NLC ; ð5Þ where CLC = kB(a, y) or k*(a, y), i.e., the lung cancer mortality (baseline) rates in the Wismut cohort or the external population rates (former GDR), respectively, NLC is the lung cancer death rate among never smokers in the group of interest (assumed to be equal to N*LC), N*LC is the lung cancer death rate among never smokers in CPS-II and S*LC is the lung cancer mortality rate for smokers in CPS-II. If the SIF is used to describe the age-specific or cumulative hazard of smoking, all calculated SIF values that exceed 1.0 (when CLC [ S*LC) are set equal to 1.0. Likewise, if SIR \ 0 (which could happen if CLC \ N*LC, that might be the case for very young age-groups), the calculated values are set to 0. SIF is a measure that ranges from 0 to 1. An SIF of 1 is equivalent to a population comprised entirely of lifetime smokers (in the reference population), and an SIF of 0 is equivalent to a population comprised entirely of never smokers. Patterns in the age-specific SIF values indicate the maturity of the smoking epidemic in a given group: if the epidemic is in its early phases, the SIF values are high in the younger age-groups and low in the older; if the epidemic is ‘mature’, the elevated SIF values extend across the entire age range; and when the epidemic is declining, the SIF values are lower in the younger agegroups than in the older ones. The Peto–Lopez method is widely used in epidemiology and social science and is not effectively challenged by the tobacco industry (http://www. deathsfromsmoking.net).

Results The absolute and cumulative number of lung cancer deaths in the full Wismut cohort (1946–2003) is shown in Fig. 1a and b as a function of calendar year from 1960 and age attained from 20 years. It can be seen from these figures that (a) the absolute number of lung cancers occurring reaches a maximum around 67 years of age, increases steadily from 1960 to 1980 and reaches a plateau thereafter; (b) the cumulative number of lung cancers has a sigmoid shape as a function of age attained, but with calendar year, there is a gradual non-linear increase up to 1970, which becomes linear thereafter. On restricting the full data range available in the cohort (1946–2003) to match the availability of external lung cancer rates for the former GDR, i.e., by omitting the cohort data from 1946 to 1959, the main epidemiological

Author's personal copy

Number of lung cancer deaths

(a)

61

120

3500

100

3000 2500

80

2000

60

1500

40

1000

20

Cumulative


500

0 1960

0 1970

1980

1990

2000

Number of lung cancer deaths

(b)

700

3500

600

3000

500

2500

400

2000

300

1500

200

1000

100

500

0

Cumulative

Calendar year

0 20

40

60

80

Attained age (years)

Fig. 1 a The number of lung cancer deaths in the Wismut cohort as a function of calendar year. The left-hand ordinate and crosses show the simple numbers, and the right-hand ordinate and line show the cumulative numbers. b The number of lung cancer deaths in the Wismut cohort as a function of age attained. The left-hand ordinate and crosses show the simple numbers, and the right-hand ordinate and line show the cumulative numbers

quantities of interest relevant to the analysis presented here are as follows: there are 3,001 lung cancers and 1.76 million person-years. The percentage of missing causes of deaths, based on 19,501 known causes of death and 1,183 missing causes of death, is 5.7%. A disproportionate number of the total missing causes of death (i.e., 494 of the total 1,183) occurred during the period 1960 to 1969. This is due to the late start of data collection for this cohort on 1.1.1999, linked with the fact that death certificates were rarely kept by the authorities for more than 30 years. Qualitative age, period and cohort analysis Qualitative differences between GDR external rates and internal birth cohort baseline rates were based on the latter obtained directly from the strata associated with a previously published (Walsh et al. 2010b) excess risk model (Eqs. 3 and 4), with the previously published parameters also given and described here in Table 1. This model is linear in radon exposure with exponential effect modifiers that depend on age at median exposure, time since median exposure and radon exposure rate. In this model, the central estimate of ERR/WLM is 1.06% (95% CI: 0.69%; 1.42%) for an age at median exposure of 33 years, a time since median exposure of 11 years and an exposure rate of

Table 1 ERR model for lung cancer and radon exposure, with exponential radon exposure rate, age and time effect modifiers (Eqs. 3 and 4) Model description

Parameter name

Fitted value

f(w,er) g(e, t)

bw

1.06 (0.69; 1.42)

exp(ww)

0.95 (0.93; 0.96)

exp(10a) exp(10e)

0.68 (0.57; 0.82) 0.46 (0.39; 0.55)

The ERR fit parameters (i.e., the b‘s) are in units of ERR per 100 WLM, with model centering, where applicable, at an age at median exposure of 33 years (relevant parameter is exp(10a)), a time since median exposure of 11 years (relevant parameter is exp(10e)), and an exposure rate of 2.7 WL (relevant parameter is exp(ww)). All fit parameters are quoted with 95% Wald-type confidence intervals

2.7 WL. This central ERR decreases by 5% for each unit of exposure rate increase. The ERR decreases by 32% with each decade increase in age at median exposure and also decreases by 54% with each decade increase in time since median exposure. The results are presented in Fig. 2a (external rates) & b (internal strata) for the time period effects, and in Fig. 2c (external rates) & d (internal strata) for the birth cohort effects. A direct comparison of Fig. 2a and b shows that whereas the external rates tend to decline beyond age 70 years with increasing age in most time periods considered, the internal rates do not and are generally higher. A similar effect can also be seen by comparing Fig. 2c and d, for different birth cohorts. The earliest birth cohort available for the internal cohort is 1900–1904, and it can be seen from Fig. 2d that the variation in the death rates between attained-age categories is much greater for this group than for the later birth cohorts. This effect is caused by the relatively fewer person-years of observation available for this group. Results for smoking impact factors from the Peto–Lopez indirect method Estimates of the fraction of tobacco-attributable deaths from the internal and external baseline rates were obtained by the indirect method (proposed by Peto et al. 1992) described above. The overall age-group-specific lung cancer rates for both the Wismut internal baseline and the former GDR are shown in Fig. 3 along with the lung cancer rates among smokers and non-smokers in the large prospective cohort study conducted by the American Cancer Society CPS-II study from the 1980s mentioned above. The age-specific SIFs are shown in Fig. 4 and indicate that the impact of smoking is greater in the Wismut baseline group for older ages, i.e., beyond 65 years, and lower for younger ages. Since the SIFs are decreasing with increasing age more

123


Radiat Environ Biophys (2011) 50:57–66 b Fig. 2 a GDR external rates: The number of lung cancer deaths per

Number of lung cancer deaths per 100000 persons

(a) 1000 800 600

Period 1965-69

Period 1970-74

Period 1975-79

Period 1980-84

Period 1985-89

Period 1990-94

Period 1995-99

Period 2000+

400 200 0 50

55

60 65 70 75 Lower category bound of attained-age

80

85

100,000 persons as a function of age attained in different time periods. b Internal cohort baseline rates: The number of lung cancer deaths per 100,000 persons as a function of age attained in different time periods. Note the vertical scale has not been extended to include two points with high rates of about 2,000 deaths per 100,000 persons, due to the very small number of person-years contributing to these spurious values. c GDR external rates: The number of lung cancer deaths per 100,000 persons as function of age attained in different birth cohorts. d Internal cohort baseline rates: The number of lung cancer deaths per 100,000 persons as a function of age attained in different birth cohorts. Note the vertical scale has not been extended to include two points with high rates of about 2,000 deaths per 100,000 persons, due to the very small number of person-years contributing to these spurious values

1000

1600

800 600 400 200 0 50

55

60

65

70

75

80

85

Lower category bound of attained-age

Number of lung cancer deaths per 100000 persons.


(b)

1200

800

400

0

800 600

Cohort 1900-04

Cohort 1905-09

Cohort 1910-14

Cohort 1915-19

Cohort 1920-24

Cohort 1925-29

Cohort 1930-34

Cohort 1935-39

25 35 45 55 65 75 85 Lower category bound of attained-age (years)

Fig. 3 Age-specific lung cancer mortality rates comparing: CPS-II male current smokers of European descent over six years of follow-up (1982–1988), CPS-II male never smokers over 22 years of follow-up (1982–1988), Wismut cohort internal baseline rates (1960–2003) and male rates for the population of the former GDR (1960–2003). The CPS-II data were accessed directly from (Thun et al. 2008 online tables S4 and S21)

400 200 0 50

55

60

65

70

75

80

85

Lower category bound of attained-age


(d)

1000 800 600

1 Smoking Impact Factor


(c) 1000

CPS-II current smokers Wismut baseline former GDR CPS-II never smokers

0.8 0.6 0.4 Wismut baseline

0.2

former GDR 0 25

400 200

35 45 55 65 75 Lower category bound of attained-age (years)

85

Fig. 4 Age-specific smoking impact factors (SIFs) computed as described in the main text applying the rates given in Fig. 3

0 50

55

60 65 70 75 Lower category bound of attained-age

80

85

strongly in the former GDR than in the Wismut group, it can be inferred that the smoking epidemic is more mature in the latter group.

123

SMR results (comparison with external rates) The proportion of causes of death that are known, P, was also determined in each radon exposure class and for each calendar year (top panels of Figs. 5 and 6) and was found to show some non-random variation. Therefore, in the

Author's personal copy Proportion of causes of death that are known


63

1 0.9 0.8 0.7 0.6 0.5 0.4 1960

1970

1980

1990

2000

2010

2000

2010

Calendar year

Mean fine correction factor

1 0.9 0.8 0.7 0.6 0.5 0.4 1960

1970

1980

1990

Calendar year

Proportion of causes of death that are known

Fig. 5 Top panel: actual proportion of causes of death that are known in each calendar year between 1960 and 2003; Bottom panel: personyear-weighted average in each calendar year of the fine correction factor that was applied to each Poisson cell in the tabulated data

1 0.95 0.9 0.85 0.8 1

2

3

4

5

6

7

all related to calendar period, an additional adjustment to the observed number of deaths (O) made only by calendar period was tested. It can be seen from Table 2 that the number of deaths (1960–2003) observed (O) was significantly higher than expected (E) from national rates. The fitted values of SMR (SMR = O/E) given in Table 2 depend slightly on whether a correction of O (uncorrected) to O* (with a single average correction factor of 0.943, i.e., O* = O/0.943) or to O** (with a fine correction factor in each Poisson cell) was made for missing causes of death. The SMR values with 95% confidence intervals (CI) are 1.91 (1.85; 1.98), 2.03 (1.96; 2.10) or 2.06 (1.99; 2.14) for O, O* and O**, respectively. However, it can be seen from Table 2 that correction by calendar year produces results that are very similar to those obtained with the single average correction factor. Simple ERR parametric cohort risk models (comparisons with external and internal rates) Statistically significant cumulative radon exposure effects in terms of ERR/WLM values and 95% CIs are also given in Table 2. These are: relative to the external GDR rates, 0.0022 (0.0021; 0.0024), 0.0025 (0.0023; 0.0026) and 0.0025 (0.0023; 0.0026) for O, O* and O**, respectively; and relative to the internal cohort baseline rates, 0.0019 (0.0017; 0.0022), 0.0019 (0.0017; 0.0022) and 0.0018 (0.0016; 0.0021) for O, O* and O**, respectively. As in the last section, it can be seen from Table 2 that correction by calendar year produces results that are very similar to those obtained with the single average correction factor.

Radon Exposure Category

Internal to external baseline comparison

Mean fine correction factor

1 0.95 0.9 0.85 0.8 1

2

3

4

5

6

7

Radon Exposure Category

Fig. 6 Top panel: actual proportion of causes of death that are known in radon exposure category; Bottom panel: person-year-weighted average in each radon exposure category of the fine correction factor that was applied to each Poisson cell in the tabulated data

determination of SMR values, the observed number of deaths was applied: without adjustment, with adjustment for one mean cohort value of P and with an adjustment value of P obtained in each Poisson cell (bottom panels of Figs. 5 and 6). Since the Poisson cells are classified by calendar period, and age and exposure variables, which are

Statistically significant differences were found in the ratio of internal to external baseline risks. These ratios and 95% CIs are also given in Table 2. These are 1.098 (1.027; 1.169), 1.165 (1.089; 1.240) and 1.200 (1.122; 1.278) for O, O* and O**, respectively. The result pertaining to O* that the internal baseline rates were found to be greater than the external baseline rates by 16.5% has previously been reported (Walsh et al. 2010a). This result is an average value for the whole range of covariates, but a further analysis that recomputed this value for two agegroups of \70 and C70 years gives 1.097 (1.011; 1.184) and 1.34 (1.188; 1.499), respectively. This latter result that the internal baseline rates were found to be higher than the external baseline rates by 34% for ages over 70 years quantifies the qualitatively observed differences in Fig. 2a–d above. Since Fig. 5 indicates that the proportion of causes of death that are known are much lower for the time period 1960–1969 (i.e., at 0.63 with O = 187 lung cancers,

123



Table 2 Results of fitting the models for the standardized mortality ratios (SMR) given in Eqs. 1 and 2 of the main text Correction mode for missing Parameter Meaning causes of death name None

b

SMR

Fitted value

Average

b

2.03 (1.96; 2.10)

b

2.03 (1.96; 2.10)

Fine

b b1

Average

Fitted value [10-2]

1.91 (1.85; 1.98)

By calendar year None

Parameter Meaning name

2.06 (1.99; 2.14) b2

b1

internal = external fixed at 1 baseline fixed at 1

By calendar year Fine

b1 b1

fixed at 1 fixed at 1

b2 b2

0.244 (0.229; 0.260) 0.246 (0.230; 0.261)

None

b1

Average

b1

internal to external 1.098 (1.027; 1.169) b2 baseline ratio 1.165 (1.089; 1.240) b2

ERR/WLM relative to 0.192 (0.167; 0.216) internal baseline 0.192 (0.167; 0.216)

By calendar year

b1

1.163 (1.087; 1.239) b2

0.192 (0.167; 0.216)

Fine

b1

1.200 (1.122; 1.278) b2

0.183 (0.159; 0.207)

b2

ERR/WLM relative to external baseline

0.223 (0.208; 0.238) 0.245 (0.229; 0.261)

Values given in parentheses represent 95% CI intervals

O* = 187/0.63 = 297) than for later years, the internal to external baseline risk ratio was also computed for the data from 1970. Ratios and 95% confidence intervals for the data from 1970 are 1.123 (1.049; 1.198), 1.165 (1.087; 1.244) and 1.184 (1.105; 1.262) for O, O* and O**, respectively. Since there are only small differences between the ratios for the data from 1960 and the data from 1970, it can be inferred that the lower values for the proportion of causes of death that are known before 1970 have little influence on the observed differences in internal to external baseline risks.

Discussion It is generally acknowledged in the field of radiation epidemiology that a cohort that is good for investigating the detrimental effects of ionizing radiation should have a substantial proportion of unexposed cohort members and that the unexposed cohort members form a more suitable group for assessing the radiation-associated disease rates than an external population group from the same geographical area. However, not many papers have actually investigated the detailed reasons for this generally accepted preference. The work presented here is aiming at filling this gap by examining systematic differences in the lung cancer mortality rates, which were found to exist when comparing the Wismut baseline group (i.e., the lung cancer rates remaining after accounting for the strong risk factor, radon) to the general population of the former GDR. There are several other risk factors for lung cancer, which are relevant to the Wismut cohort members at exposure levels that either do not apply at all to the general population or only apply at much lower levels. Such risk factors include exposure to gamma radiation, long-lived radionuclides,

123

fine dust, arsenic dust, quartz fine dust, asbestos and diesel fumes. All of these factors could have had some influence on the observed differences in lung cancer baseline rates. The analysis presented here should be repeated in the future when all of the other risk factors and their interactions have been fully evaluated. However, it is cautiously assumed now that there are two major factors that emerge from the many possible covariables that contribute to such differences in the current baseline rates, namely differences in the characteristics of routine medical screening programs (p. 666 Runge 1999) and differences in the prevalence of smoking as discussed below. A screening program in the Wismut mining company was started in 1952, primarily because of the many observed cases of Silicosis and Tuberculosis. This was carried out initially as a pilot project, which was then extended to the general population of the former GDR in 1953/1954 with the primary aim of detecting tuberculosis. Initially, annual medical checks were carried out in the GDR that were later reduced to biennial checks (with chest X-rays for those over 40 years of age). At the Wismut company, the medical checks were supposed to be annual and independent of age and accumulated exposure. However, it is known that this regime was not strictly adhered to by the employees, since not turning up to medical examinations had no work-related consequences. It can therefore be assumed that, in the absence of symptoms of ill health, medical screening in the Wismut employees was only marginally more frequent than in the general population of the GDR. In contrast, Wismut employees with symptoms such as breathing difficulties were carefully monitored and given chest X-rays at short time intervals. Since there were about 20,000 confirmed cases of silicosis among Wismut employees up to 1990, treated by many associated medical

Author's personal copy Radiat Environ Biophys (2011) 50:57–66

specialists for this disease, it is reasonable to assume that lung cancer was diagnosed at an earlier stage in Wismut employees than in the general population of the former GDR. This would not, however, explain the observed differences in the lung cancer mortality rates, although a minor part of these differences may be due to increased detection of lung cancers in the miner cohort, due to higher autopsy rates than in the general population of the former GDR. The Wismut cohort data only include a very limited amount of smoking-related information and only for a small percentage of the subjects. Detailed smoking information has recently been collected for about 2000 cohort members and assessed in a nested case–control study that examined the influence of smoking on the radon-related lung cancer risk (Schnelzer et al. 2010). Since data on smoking are not available for the majority of Wismut cohort members, it is only possible to obtain indirect indications and apply methods for ascertaining the possible effects of smoking in this cohort. There have already been indications that the prevalence of smoking in the Wismut employees was higher than in the general population, from two case–control studies in the former GDR regions of Thuringia and Saxony. These two studies investigated the lung cancer risk due to radon and other risk factors, one concerning Wismut miners and the other excluding miners (Bru¨ske-Hohlfeld et al. 2006; Wichmann et al. 1999). Based on standardized personal interviews that included detailed questions about smoking history, only 15.2% of male controls among the Wismut miners analyzed (Bru¨skeHohlfeld et al. 2006) were classified as never having smoked, whereas the percentage in the other studies’ control group of male non-miners was 26.5% (Wichmann et al. 1999). It is also interesting to note that the former Soviet general director of the Wismut company, General Malzews, ordered additional remuneration and performance bonuses in the form of alcohol and tobacco products— ‘‘Every employee working in the mines or at ground level will receive 50 and 30 cigarettes per month, respectively. By above average performance an additional 50 cigarettes per 10 days are payable’’ (p. 780 Runge 1999). So, it was possible for Wismut employees to earn up to 200 cigarettes per month in addition to any they may have purchased—a large quantity by past and current standard—which can be compared to the average number of cigarettes consumed per month in the former GDR, which rose from 89 in 1960 to 154 in 1989 (Statistisches Bundesamt 2005). It can therefore be assumed that not only the smoking prevalence and the amount of tobacco consumed were higher, but also the age at the start of smoking was probably younger, than in the general population.

65

These earlier indications and the socially based evidence are corroborated by the analysis presented here in terms of the SIFs from the Peto–Lopez indirect method, which indicates that the smoking epidemic is more mature in the Wismut miners than in the general population of the GDR. A possible criticism of the SIF method could be that the lung cancer rates among lifelong non-smokers could have changed over time and may be different between countries. However, a recent analysis of 13 cohorts and 22 cancer registry studies (Thun et al. 2008) found no indication either that lung cancer rates have changed among never smokers in the age range 40–69 years in the US since the 1930s or that death rates have changed appreciably among never smokers from CPS-I (1959–1972) to CPS-II (1982–2004), where these latter two cohorts are the only cohorts currently available for assessing this point. The calculation of the SIFs involves the rates for current smokers; these rates were obtained over six years of follow-up of CPS-II (1982–1988) for men of European descent where this time period falls approximately in the middle of the time period for the Wismut cohort follow-up and GDR rates (1960–2003). The analyses for the age category–specific SIFs presented above were repeated for Wismut and GDR rates that were restricted to the 1982–1988 time period, but this restriction did not affect the main results to any notable degree. Other limitations to the SIF analysis might exist because factors relevant to the determination of lung cancer risk may have differed between the ACS cohort and the population and cohort studied here. Such factors include the daily amount and number of years smoked, exposure to second-hand smoke, brand of cigarette, age at initiation and inhalation habits. Methods applied here for adjusting for missing causes of death associated with the data collection process for the Wismut cohort only indicated a minor effect of missing causes of death on both the differences in internal and external baseline rates and on the radon-related risk per unit of exposure. This is related to the fact that there are only about 6% of total deaths and 6% of total lung cancer deaths occurring in the period with the lowest proportion of causes of death that are known, i.e., between 1960 and 1970.

Conclusions This work has shown that systematic differences in the lung cancer mortality rates exist when comparing the Wismut baseline (i.e., the lung cancer rates remaining after accounting for the strong risk factor, radon) to the lung cancer mortality of the general population of the former GDR. For this reason, it is generally recommended to apply the Wismut internal rates in preference to the external

123


mortality rates, for baseline lung cancer rate determination, in future detailed risk models of radiation-associated lung cancers in this cohort. Even though the missing causes of death associated with the data collection process for the Wismut cohort show some non-random variation with calendar years and radon exposure category, they were found to have only a minor effect on both the differences in internal and external baseline rates, and on the radon-related risk per unit of exposure. The Peto–Lopez method has been applied here and has indicated that the systematic differences in the lung cancer mortality rates between the internal and external baselines could be due to differences in the maturity of the smoking epidemic. In view of the current limitations, it is recommended that the analysis presented here should be repeated in the future and based on either a single or model-averaged preferred model that includes all of the other risk factors and their interactions. At this point in time, however, such risk factors have not yet been fully evaluated for the Wismut cohort. Acknowledgments The German Federal Commissioner for Data Protection and Freedom of Information has issued a special approval for this research, which constitutes an exemption from the necessity to obtain human subjects approvals. This work was partially funded by the EU Alpha-Risk project and by the Federal Ministry of Education and Research (BMBF), Germany (Competence Network Radiation Research). The authors thank the German Federation of Institutions for Statutory Accident Insurance and Prevention (DGUV) and the Miners’ Occupational Compensation Board (Bergbau Berufsgenossenschaft) for their continuous support over many years. The field work for the follow-up was conducted by I ? G Gesundheitsforschung and Mediveritas GmbH. Their commitment helped to achieve the low percentage of lost to follow-up. We also thank the members of the Wismut Working Group of the German Radiation Protection Commission for their continued advice. Special thanks are due to Prof. Werner Ru¨hm for a valuable initial discussion and some very helpful suggestions.

References Bru¨ske-Hohlfeld I, Rosario AS, Wolke G, Heinrich J, Kreuzer M, Kreienbrock L, Wichmann HE (2006) Lung cancer risk among former uranium miners of the WISMUT Company in Germany. Health Phys 90:208–216 Dahmann D, Bauer HD, Stoyke G (2008) Retrospective exposure assessment for respirable and inhalable dust, crystalline silica and arsenic in the former German uranium mines of SAG/SDAG Wismut. Int Arch Occup Environ Health 81:949–958 Kreuzer M, Walsh L, Schnelzer M, Tschense A, Grosche B (2008) Radon and risk of extra-pulmonary cancers—Results of the German uranium miners cohort study, 1960–2003. Br J Cancer 99:1946–1953

123

Radiat Environ Biophys (2011) 50:57–66 Kreuzer M, Schnelzer M, Tschense A, Walsh L, Grosche B (2009a) Cohort profile: the German uranium miner cohort study (WISMUT cohort), 1946–2003. Int J Epidemiol. doi:10.1093/ije/ dyp216 Kreuzer M, Grosche B, Schnelzer M, Tschense A, Dufey F, Walsh L (2009b) Radon and risk of death from cancer and cardiovascular diseases in the German uranium miners cohort study: follow-up 1946–2003. Radiat Environ Biophys 49(2):177–185 Lehmann F, Hambeck L, Linkert KH, Lutze H, Meyer H, Reiber H, Renner HK, Reinisch A, Seifert T, Wolf F (1998) Belastung durch ionisierende Strahlung im Uranerzbergbau der ehemaligen DDR. Hauptverband der gewerblichen Berufsgenossenschaften, St. Augustin Peto R, Lopez AD, Boreham J, Thun M, Heath C Jr (1992) Mortality from tobacco in developed countries: indirect estimation from national vital statistics. Lancet 339:1268–1278 Peto R, Lopez AD, Boreham J, Thun MJ, Heath C Jr (1994) Mortality from smoking in developed countries, 1950–2000: indirect estimates from national vital statistics. Oxford University Press, Oxford. (Second edition available online at http://www.ctsu. ox.ac.uk/*tobacco/) Powles J (2000) The Peto/Lopez indirect method for estimating the accumulated mortality hazard caused by smoking (GBD2000 protocol). Available from: http://www.phpc.cam.ac.uk/mst/phtop/ peto_lopez_method_gbd2000.doc. Accessed 1 March 2010 Preston DL, Lubin JH, Pierce DA, McConney ME (1993) Epicure user’s guide. HiroSoft, Seattle Rittgen W, Becker N (2000) SMR Analysis of historical follow-up studies with missing death certificates. Biometrics 56:1164–1169 Runge W (editor in chief) (1999) Chronik der WISMUT, Wismut GmbH (2738 pages, in German) Schnelzer M, Kreuzer M, Tschense A, Hammer G, Grosche B (2010) Accounting for smoking in the radon related lung cancer risk among German uranium miners: results of a nested case-control study. Health Phys 98:20–28 Statistisches Bundesamt (2005) Fachserie 14: Finanzen und Steuern, Reihe 9.1.1 Absatz von Tabakwaren, Abbildung 2.5.2: Jahrliche Pro-Kopf-Verbrauch (in Stu¨ck) an Zigaretten in Deutschland 1955 bis 2004 Thun MJ, Hannan LM, Adams-Campbell LL, Boffetta P, Buring JE, Feskanich D, Flanders WD, Jee SH, Katanode K, Kolonel LN, Lee IM, Marugame T, Palmer JR, Riboli E, Sobue T, Avila-Tang E, Wilkens LR, Samet JM (2008) Lung cancer occurrence in never-smokers: an analysis of 13 cohorts and 22 cancer registry studies. PLoS Med 5(9):1357–1371 Walsh L, Tschense A, Schnelzer M, Dufey F, Grosche B, Kreuzer M (2010a) The Influence of radon exposure on lung cancer mortality in German uranium miners, 1946–2003. Radiat Res 173:79–90 Walsh L, Dufey F, Tschense A, Schnelzer M, Grosche B, Kreuzer M (2010b) Radon and the risk of cancer mortality—Internal Poisson models for the German uranium miners cohort. Health Phys 99(3):292–300 WHO (1992) International statistical classification of diseases and health related problems: the ICD-10 (German edition 1.3.1999), Geneva Wichmann HE, Gerken M, Wellmann J, Kreuzer M, Kreienbrock L, Keller G, Wo¨lke G, Heinrich J (1999) Lungenkrebsrisiko durch Radon in der Bundesrepublik Deutschland (Ost)- Thuringen und Sachsen. ecomed, Landsberg. (in German)