3.1. Relative Risks in Case-Control Studies
In both cohorts, a total of 1,542 cases of lung cancer were observed by the end of 2010 (1,249 among miners and 293 in the residential study). Smoking data were collected for 1,029 and 289 cases and 2,648 and 1,156 controls, in the occupational and residential nested studies, respectively, see . In the present study, all analyses were carried out only for subjects with smoking information.
Summary of nested case-control studies.
Smoking categories in the present studies were considered as follows: never-smokers, ex-smokers (who quitted before 10 years or more) and other smokers including current smokers and ex-smokers who quitted before 10 years or less. These categories follow findings from preliminary risk calculations which showed that the relative risks among current smokers and ex-smokers (<10 years) were very similar. Numbers of cases and controls by smoking categories are given in .
Summary of cases and controls by smoking.
Relative risks of lung cancer in relation to smoking and radon exposure categories (in terms of odds ratios) are given in . Here, odds ratios are given separately (crude OR) for smoking categories (ignoring radon exposure) and for categories of radon exposure (ignoring smoking). The risks are estimated also for categories of smoking with the adjustment to radon categories and similarly for radon categories adjusted to smoking (adjusted OR).
Numbers of cases and controls and odds ratios in categories by smoking and radon exposure.
Smoking adjusted odds ratios in categories of radon exposure (in terms of multiplicative model where OR(smoking, radon) = OR(smoking) × OR(radon)) did not substantially change the effect from radon when smoking was ignored (). However, when the linear effect of cumulated radon exposure was evaluated separately for never- and ever-smokers, the risk coefficients (excess relative risk per unit exposure) were substantially (5 times) higher in never-smokers both in the occupational and residential studies (). The difference is significant in the occupational study (p = 0.033), but not in the residential study as the statistical power in this study is lower due to smaller numbers of cases and much lower cumulated exposures. In terms of radon progeny, mean exposure among miners was 69 MBq·m−3·h and among residents 30 MBq·m−3·h (assuming F = 0.4, 7,000 h at home annually). For the highest exposure categories, 300 WLM correspond to 189 MBq·m−3·h and 600 Bq·m−3 correspond to 50 MBq·m−3·h. In terms of intake (assuming breathing rate 1.2 m3/h in miners and 0.8 m3/h in residents) exposure 300 WLM corresponds to 227 MBq and exposure 600 Bq·m−3 corresponds to 40 MBq.
Coefficients of excess relative risk per unit exposure by smoking categories
Graphically, the dependence of the relative risks on cumulated exposure by smoking categories is given in . 90%CI calculated using the method of floating risk by Plummer [12
], RR scales adjusted for RR = 1 at zero exposure among never-smokers. 1 WLM corresponds to 629 kBq·m−3
·h (Rn progeny) 30-year exposure at 1 Bq·m−3
corresponds to 84 kBq·m−3
·h (Rn progeny), assuming F = 0.4 and 7,000 h at home annually.
Relative risks (RR) by cumulated radon exposure and smoking categories. Panel (a) occupational study, panel; (b) residential study, including the multiplicative (dotted thin lines) and the best geometric mixture models (full thin lines).
3.2. Geometric Mixture Models
The above observed differences in the exposure-response relationship in smoking categories lead to the evaluation whether the combined effect of radon and smoking is additive or multiplicative. This issue is solved by the geometric mixture models with an additional parameter (θ), which include the additive (θ = 0) and multiplicative (θ = 1) models as special cases. The best model is the one with the lowest deviance as graphically presented in .
Dependence of deviances on mixing parameter θ in the occupational study for models G1 and G2 (panel a), and in the residential study for model G1 (panel b).
The results of fitting for different mixing parameters are given in . The best fit is achieved for θ = 0.2 in the occupational study (model G2) and for θ = 0.3 in the residential study (model G1). Both these values suggest that the best model of the combined effect from radon and smoking is closer to the additive model than to the multiplicative model.
Parameters in fitted geometric mixture models is the occupational and residential studies for selected parameters θ.
The dependence of the risk on residential exposure in (b) is also given in terms of the fitted multiplicative and geometric mixture models. The main difference between the two models is in the estimates of risks related to smoking at zero exposure. The risks from smoking at zero exposure are substantially lower in the multiplicative model in comparison to the additive or the best geometric mixture models. The most likely reason is that the induced lung cancer cases from radon exposure do not contribute to spontaneous cases in the proportion usually observed in the smoking and non-smoking populations. In fact, the contribution from radon exposure is about the same in smokers and non-smokers, which corresponds more to the additive model.
The θ confidence interval for the occupational study corresponding to the G2 model is 0.05–0.48, suggesting a substantial difference from the multiplicative model. A θ confidence interval in the residential study is much wider because of lower numbers of cases and lower exposures and covers the entire interval 0–1. It should be noted that in the occupational study, the fit in terms of G2 model is substantially better than the G1 model (chi-sq = 24.1, p < 0.001).
3.3. Estimation of Induced Lifetime Risk from Indoor Radon
The resulting estimates from indoor radon in geometric mixture models are translated into estimates of lifetime risks in the general population of the Czech Republic. Lifetime risks are calculated using mortality statistics in 2010 and estimated proportion of smokers (57% among men, 43% among women) in the country [13
]. The lung cancer mortality rates in never-smokers are taken from [13
] and the hypothetical mortality in smokers is calculated using the general mortality from lung cancer and average smoking prevalence (50%) in the Czech population, see . The calculations of lifetime risk from lung cancer are based on life-tables in the Czech Republic in 2010 [14
]. Lung cancer age specific rates in “non-exposed” population are taken according to the Czech population data in 2010 [13
] and lung cancer rates in exposed population are calculated according to the multiplicative and geometric mixture models, .
Figure 3 Age specific lung cancer rates (per 100,000) in the Czech Republic (men and women combined) and estimates of rates among smokers and non-smokers using mean prevalence (50%) of smokers and estimates in non-smokers according to Peto et al. .
Age specific lung cancer rates (per 100,000) in a hypothetical populations of never- and ever-smokers exposed to 100 Bq·m−3 according to the resulting geometric mixture model (●) and the multiplicative model (▲).
Results in terms of lifetime lung cancer numbers in 100,000 population are given in . As expected, numbers of lung cancers in smokers exceed numbers in non-smokers several times—about 14 times in non-exposed population and in exposed population according to the multiplicative model. However, the ratio of smoking to non-smoking lung cancers in exposed population is only about 10 according to the resulting geometric mixture model. Substantial difference between the two models is in numbers estimated in non-smokers—by a factor of nearly 5.
Lung cancer lifetime risks (per 100,000) according to the multiplicative and best geometric mixture models in smoking and non-smoking populations (according to Czech statistics in 2010).