Residential radon and lung cancer—detailed results of a collaborative analysis of individual data on 7148 persons with lung cancer and 14 208 persons without lung cancer from 13 epidemiologic studies in Europe

OBJECTIVES
Studies seeking direct estimates of the lung cancer risk associated with residential radon exposure lasting several decades have been conducted in many European countries. Individually these studies have not been large enough to assess moderate risks reliably. Therefore data from all 13 European studies of residential radon and lung cancer satisfying certain prespecified criteria have been brought together and analyzed.


METHODS
Data were available for 7148 persons with lung cancer and 14,208 controls, all with individual smoking histories and residential radon histories determined by long-term radon gas measurements.


RESULTS
The excess relative risk of lung cancer per 100 Bq/m3 increase in the observed radon concentration was 0.08 [95% confidence interval (95% CI) 0.03-0.16; P=0.0007] after control for confounding. The dose-response relationship was linear with no evidence of a threshold, and it remained significant when only persons with observed radon concentrations of <200 Bq/m3 were included. There was no evidence that the excess relative risk varied with age, sex, or smoking history. Removing the bias induced by random uncertainties related to radon exposure assessment increased the excess relative risk of lung cancer to 0.16 (95% CI 0.05-0.31) per 100 Bq/m3. With this correction, estimated risks at 0, 100, and 400 Bq/m3, relative to lifelong nonsmokers with no radon exposure, were 1.0, 1.2, and 1.6 for lifelong nonsmokers and 25.8, 29.9, and 42.3 for continuing smokers of 15-24 cigarettes/day.


CONCLUSIONS
These data provide firm evidence that residential radon acts as a cause of lung cancer in the general population. They provide a solid basis for the formulation of policies with which to manage risk from radon and reduce deaths from the most common fatal cancer in Europe.


Overall results and comparison with other studies of residential radon and lung cancer
This analysis combined data from the 13 European studies of residential radon and lung cancer that have included at least 150 lung cancer cases selected according to clear rules, together with controls representative of the populations from which the lung cancer cases had been drawn, and in which individual information on both smoking history and radon exposure history, based on long-term measurements of radon gas concentrations, was available in the detail that was agreed upon when collaboration began. Other European studies of residential radon and lung cancer that do not meet these criteria have been reported by Axelson (54) and Conrady et al (55). The data presented here provide, for the first time, very strong evidence (P=0.0007), based on individual data, of an association between residential radon concentration during the previous 35 years and the risk of lung cancer after adjustment for smoking history. There was no evidence of heterogeneity between the different studies in this relationship (P=0.94), nor were the results dominated by any individual study. In addition, there was no evidence that the estimated doseresponse relationship depended on detailed aspects of the study design (table 12) or the characteristics of the radon measurements (table 13).
The estimated excess relative risk of lung cancer of 0.084 (95% CI 0.030-0.158) per 100 Bq/m 3 increase in the TWA observed radon concentration is statistically consistent with the estimate of 0.11 (95% CI 0.00-0. 28) found in a recent collaborative analysis of seven North American studies of residential radon and lung cancer (56). Compared with the North American collaborative analysis, data from larger numbers of persons were available for analysis in the European study (7148 lung cancer cases and 14 208 controls, against 3662 lung cancer cases and 4966 controls), and the TWA observed radon concentrations tended to be higher (10.4% of the persons with >200 Bq/m 3 , compared with 4.9% in the North American collaborative analysis), and these differences are likely to account for the greater precision of the estimate in the European study. The results of the present European study are also consistent with those of a recent study carried out in Gansu, China (57), which found an excess relative risk of 0.19 (95% CI 0.05-0. 47) per 100 Bq/m 3 observed radon concentration on the basis of 768 lung cancer cases and 1659 controls, although they differed substantially from the findings of an earlier Chinese study carried out in Shenyang, which reported a negative dose-response relationship [the excess relative risk at 100 Bq/m 3 being -0.16 (95% CI -0.2--0.1) on the basis of 308 lung cancer cases and 356 controls (58,59)]. One possible explanation for this discrepancy may be differences in the etiology of lung cancer between Europe and China and within China, as indoor air pollution is known to play a substantial role in causing lung cancer in some areas of China, including Shenyang (60). Another possible explanation may be the fact that the Shenyang study included two controls whose radon measurements were more than 50% greater than the next largest value in that study. When included, these two persons are highly influential, but when they are omitted the results from the Shenyang and Gansu studies do not differ significantly (61).
Prior to this study, meta-analyses based on published reports of studies of residential radon and lung cancer have reported statistically significant associations between radon and lung cancer (2,59,62,63). In all of these analyses, however, there was strong evidence of heterogeneity in the dose-response relationships observed in the different studies. The main explanation for this finding is likely to be that the degree to which the individual studies corrected for confounding by smoking differed between the different studies, and none had sufficient data to be able to correct in as much detail as in the present study. Other explanations for the heterogeneity in the meta-analyses based on published data may include differences in the statistical methods used and, possibly, also the inclusion of the Shenyang study. One other meta-analysis of published data (64) did not test for heterogeneity, but reported that there was some evidence of a U-shaped dose-response relationship, with a lower risk of lung cancer at 25-50 and 50-75 Bq/m 3 . However, there was no suggestion of such an effect in the Collaborative Analysis of individual data.

Shape of the dose-response, evidence at <200 Bq/m 3 and the effect of exposure at different times in the past
In the analyses presented in this report, the relationship between the odds of developing lung cancer and the Discussion TWA observed radon concentration has been modeled. The results given are based on linear models, although both linear and log-linear models provided equally good fits to the data (table 15). Reasons for preferring the linear to the log-linear model include the fact that the weight of scientific evidence suggests that a linear nonthreshold model is the most plausible dose-response relationship for ionizing radiation (65). In addition, linear rather than log-linear models have been used in other analyses of the effects of exposure to radon, including an analysis of 11 cohort studies of miners of uranium and other igneous rocks who were occupationally exposed (2,66), albeit at radon concentrations that were usually much higher than those normally found in dwellings. The ability of the linear model to summarize the data was not improved by the addition of a quadratic term (table 15), nor by the inclusion of additional categorical terms for radon, and no departure from linearity was found when the data were subdivided into categories of radon (figure 2). In particular, there was no evidence of any departure from linearity that could be attributed to the existence of radio-sensitive subpopulations, adaptive responses, or bystander effects (2,(67)(68)(69); neither was there any evidence of a protective effect of exposure to radon at low concentrations, which has sometimes been postulated (70), nor of a threshold, and, indeed, postulated thresholds of >150 Bq/m 3 lie outside the upper 95% confidence limit.
The evidence in favor of an association between residential radon and lung cancer risk did not rely on the persons with unusually high radon concentrations, and, when the linear model was fitted to the data from only the persons with TWA observed radon concentrations below a certain value, the association remained statistically significant even when only the persons with radon concentrations of <200 Bq/m 3 were considered. The estimated excess relative risk of lung cancer per 100 Bq/m 3 when only the persons with observed radon concentrations of <200 Bq/m 3 were considered was 0.140 (95% CI 0.004-0.309), in good agreement with the estimate based on the entire data set (excess RR 0.084, 95% CI 0.030-0.158).
The distribution of the residential radon concentrations is usually highly skewed and, in this analysis, the distribution of the radon concentrations in the component studies and in the overall data had long upper tails (figure C1 in appendix C). Estimates of linear trend are known to be sensitive to outlying observations and, consequently, in some of the component studies, a very small number of persons with very high radon concentrations had a disproportionate influence on the estimated excess relative risk of lung cancer per 100 Bq/m 3 .
[See table D1 in appendix D, Heid et al (71), and Schaffrath Rosario et al (72) for further discussion of this issue.] For the Collaborative Analysis, however, the amount of data available was sufficient to overcome this problem.
Analyses of studies of underground miners have suggested that radon exposures in the relatively recent past have substantially greater influence on lung cancer risk than exposure in the more distant past, with periods 5-14, 15-24, and 25-34 years previously having influences approximately in proportions of 1.00 : 0.75 : 0.50, respectively (2,66). In the present data set, an estimate of the excess relative risk per 100 Bq/m 3 weighting these three periods in proportions of 1.00 : 0.75 : 0.50 is identical to the estimate derived by giving equal weights to every year in the 30-year period ending 5 years previously, and separate estimates, each considering only exposures in the three periods 5-14, 15-24, and 25-34 years previously, were also all very similar (table 14). This difference is not surprising, for the exposures of the miners would often have varied substantially from year to year as mine ventilation was introduced, and the overall length of employment of the men in a job involving radon exposure was usually relatively short (average 5.7 years). In this study, in contrast, the radon exposures varied little from year to year because the people moved their residence only seldom, an average of only 2.7 addresses being reported during the entire 30-year period of interest. In consequence, the TWA observed radon concentrations during the three time periods were highly correlated (correlations be- . Consequently, with these data, distinguishing between the effects of exposures received in different periods of the past is impossible. An analysis estimating separately the effects of exposures received during the periods 5-19 and 20-34 years previously has recently been carried out in the Czech Republic cohort study from which the Czech Republic case-control data included in this Collaborative Analysis have been derived (73). The estimated effect of the exposures received in the distant past was lower than that for exposures received more recently. However, the confidence intervals associated with both estimates were very wide [estimated excess relative risks per 1000 Bq/m 3 a: 0.045 (90% CI 0.000-0.153) and 0.013 (90% CI 0.000-0.123), respectively].

Confounding
Residential radon concentrations vary geographically, and it was to be expected that there would be substantial confounding by study, as the studies were carried out in different geographic areas and with different case-control ratios. The overall mean TWA observed radon concentrations for all of the lung cancer cases in the study was 104 Bq/m 3 , slightly lower than the corresponding value for the controls, which was 105 Bq/m 3 , and this finding is reflected in the slight negative trend in the risk of lung cancer with increasing radon concentration when no confounding factors were taken into account (table 9, stratification 1). However, when the mean TWA observed radon concentrations was calculated separately for each study, it was larger for the lung cancer cases than for the controls, in some cases by considerable amounts, for 10 of the 13 studies (Austria, Czech Republic, Finland southern, Finland nationwide, France, Germany eastern, Italy, Sweden nationwide, Sweden never-smokers, United Kingdom). Only in three studies (Germany western, Spain, Sweden Stockholm) was the mean observed radon concentration lower for the lung cancer cases than for the controls. [See table C4 in appendix C for details.] When a weighted mean observed TWA radon concentration was calculated for the controls, with weights proportional to the numbers of lung cancer cases in each study, its value was 97 Bq/m 3 ; and a highly significant difference between the two groups of persons was revealed (P=0.0002) (table 6).
Most of the studies were matched for age and sex and, therefore, ensured that there was little confounding with these factors, and several studies were matched for region of residence. However, the ratio of cases to controls varied between the different studies, and the change from the crude excess relative risk of lung cancer per 100 Bq/m 3 of -0.005 (95% CI -0.017-0.13) to 0.023 (95% CI -0.005-0.061) when study, age, sex, and region were taken into account (table 9 stratifications 1  and 3) is primarily the effect of this correction.
In addition, there was also substantial negative confounding by smoking, and, after stratification by smoking habits, the preceding estimate of 0.023 increased substantially, to 0.084. This topic is discussed further in the section Radon and Smoking, on page 49.

Effect modification according to characteristics of the cases and controls
Overall there was little evidence of any modification of the effect of exposure to the TWA observed residential radon concentration on the relative risk of lung cancer according to the characteristics of the cases and controls. The estimated excess relative risk per 100 Bq/m 3 observed radon concentration was larger for the men than for the women (table 18), but the position was reversed in the North American studies of residential radon and lung cancer, with the excess relative risk being greater for the women than for the men (56,74). In neither study was the difference between the men and women significant statistically, and it seems likely that, in both studies, the difference between the two sexes is due to chance. There was little evidence that the effect of residential radon varied with age, nor was there evidence of such variation in the North American data. This finding is in marked contrast to the results of the analysis of the entire group of occupationally exposed miners, among whom there was a statistically significant trend in the excess relative risk per unit radon concentration, with estimated values at ages <55, 55-64, 65-74, and ≥75 years in the proportions 1.00 : 0.57 : 0.34 : 0.28, respectively (2,59,63). In the miners' studies, persons who reached the older age groups during the period of follow-up were likely to have started smoking later in life than those who were younger at the end of the follow-up. Thus, as no adjustment for confounding by smoking was carried out in the main analyses of the miners' studies, there is scope for differential confounding with the effect of smoking in the different age groups in the miners' studies. An alternative explanation might be differences in the accuracy of the exposure assessment in the different age groups. In the miners' studies, persons who reached the older age groups during the period of follow-up were likely to have started their exposure earlier in calendar time than the younger persons. In most mines, the accuracy of exposure assessment was lowest for the earliest calendar periods. Radon concentrations would also have been highest during the earliest calendar periods, and it is noteworthy that, when the analysis of radon-exposed miners was restricted to those exposed at low radon concentrations, there was no trend with age (excess relative risk per unit radon concentration, with estimated values at ages <55, 55-64, ≥65 years, respectively, in the proportions 1.00 : 0.92 : 1.43) (75). In contrast, in the present study, there was a tendency for older people to spend a greater proportion of their time indoors at home, and, among those for whom this variable was known, the percentages spending >75% of their time at home were 11.1%, 15.6%, and 24.6% for those aged <55, 55-64, and ≥65 years, respectively. As a consequence, the observed radon concentrations may be a somewhat poorer reflection of the bronchial dose due to radon and its decay products for older persons than for younger ones in the miners' studies and a somewhat better reflection of it in the residential studies.
In the present study the estimated excess relative risk of lung cancer per 100 Bq/m 3 observed radon concentration was slightly higher for the ex-smokers than for the current smokers [excess RR 0.082 (95% CI 0.003-0.211) versus excess RR 0.070 (95% CI -0.014-0.217] (table 18) and slightly higher again among the lifelong nonsmokers [excess RR 0.106 (95% CI 0.003-0.280)], although the heterogeneity between these estimates was not statistically significant. Similarly, there was no significant difference in the effect of radon exposure by smoking status in the North American data. Further discussion of this issue is given in the section Radon and Smoking, on page 49. There was evidence that the excess relative risk per 100 Bq/m 3 observed radon concentration was higher among those who had lived in a rural area for the entire 30-year period of interest than among those who had lived in an urban area for part of the time (P=0.01) (table 18). Residence in an urban, as opposed to a rural, area has not been suggested as an effect modifier of radon by any previous study. Furthermore, the effect was not consistent across all of the studies. The significance level of the effect was not extreme, and, therefore, it may be a chance finding. Alternatively, it could reflect some, as yet, unidentified difference in behavior between people living in urban and rural areas. The persons living in rural areas tended to be exposed to higher radon concentrations than those living in urban areas, and they had moved their residence less often than those living in urban areas, but no other specific difference could be identified.
There was also some evidence that the excess relative risk per 100 Bq/m 3 observed radon concentration was higher among those who usually slept with the bedroom window closed rather than open (P for difference 0.03) (table 18). The apparently modifying effect of sleeping with the bedroom window open was discussed in the Sweden nationwide study (18). As a rule, the radon concentration decreases when a window is kept open, sometimes by 50% to 70%. In the Sweden nationwide study, most of the cases and controls had slept with their bedroom window closed. All of the lung cancer cases and half the controls in the present study had died by the time the radon measurements were obtained, and, although no information on window position was obtained when the measurements were made, it is likely that most of the measurements were made with the windows closed. Therefore a greater degree of exposure misclassification would be expected for the persons who slept with their window open. The weaker association for those who slept with an open window might also be explained by greater residual negative confounding. It is likely that there is some residual confounding (eg, with cigarette smoking or other lifestyle factors) in these data, and it is possible that the confounding is stronger for those who slept with an open window. The position of the bedroom window at night as a possible effect modifier was first suggested by the data from the Sweden nationwide study, and, when this study is excluded, the effect is no longer significant statistically. Neither does it remain significant when fitted simultaneously with residence in an urban or a rural area (table 20). Thus the evidence suggesting that it is an effect modifier is not strong.
When lifelong nonsmokers were considered separately, there was no evidence that the effect of exposure differed according to whether or not the person had been married to a smoker (table 21). Neither was there any evidence that the effect of residential radon differed according to most of the other characteristics that were examined. The one exception was employment in an occupation known to be associated with lung cancer in other studies, with a larger excess relative risk of lung cancer for those who had been employed in such an occupation. However, there was no evidence in these data that such employment per se increased lung cancer risk (table 23), nor that it modified the effect of residential radon concentration when all of the data in the Collaborative Analysis were considered together after stratification for smoking history (table 18). Furthermore, the significance of the effect was primarily due to a deficit of persons with low radon concentrations having such employment. Overall, therefore, the evidence that employment in an occupation known to be associated with a risk of lung cancer is an effect modifier is not strong.

Histological type of lung cancer
The present study provides evidence of a stronger association between exposure to residential radon and smallcell lung cancer than between radon and other histological types of the disease. Although a central pathological review was carried out in only three studies (Germany eastern, Germany western, and United Kingdom), several studies contributed to the finding, and it is therefore unlikely to be an artifact of variations in the pathological classification. The result is in line with early clinical and autopsy reports on highly exposed miners in the Schneeberg region (76), where an excess of intrathoracic neoplasms was identified and the tumors were classified as lymphosarcomas, with the implication that they were mostly small-cell lung cancers. In addition, autopsy studies from highly exposed uranium miners in Jáchymov and Schneeberg in the early part of the 20th century suggested a preponderance of small-cell lung cancers (77,78). More recently, histological studies of radon-exposed uranium miners who died of lung cancer in the Czech Republic (79,80), the United States (81), and China (82) have also found that radon-induced lung cancers are more likely to be of the small-cell type than of other histological types.
In the United States study (81), a comparison was made of the localization of tumors within the lung between radon-exposed miners and nonminers, and it was found that the proportion of lung tumors and, especially, small-cell lung tumors in the central zone was greater for the miners than for the nonminers. Thus the explanation for the preponderance of small-cell lung cancers may be the fact that radon and its decay products deliver the highest radiation dose to the central zone of the lung (2). Other factors may also play a role, however, as histological studies of lung cancers in survivors of the atomic bombings of Hiroshima and Nagasaki who were exposed to uniform whole-body radiation, predominantly from gamma rays (83), have also suggested that radiation-induced lung cancers are more likely to be of the small-cell type than of other histological types. For the atomic bomb survivors, the reason for the greater risk of small-cell tumors has not yet been resolved.
Although it had been expected a priori that smallcell lung cancers would be more closely associated with radon than other cell types, the lack of any appreciable association between radon exposure and other types of lung cancer observed in the present study was not expected. In studies of occupationally exposed miners, increased risks of both squamous-cell carcinomas (81) and adenocarcinomas (82) have been found to be associated with radon exposure. Most recently, in a large study of miners in eastern Germany (84), the data suggested that all cell types were associated with radon exposure, but that high radon exposure tended to increase the proportion of both small-cell and squamous-cell cancers. A recent review has also pointed out that the histological type of lung cancer has not proved to be a definitive indicator of radon progeny as a cause of the lung cancer in radon-exposed underground miners (2).

Effects of uncertainties in the assessment of radon concentrations
Correction for the effects of random uncertainties in the assessment of radon concentrations had a major impact, with the estimated increase in relative risk per 100 Bq/m 3 nearly doubling, from 0.084 to 0.16, and the width of the associated 95% confidence interval also increasing, from 0.030-0.158 to 0.05-0.31. Data on the degree of variability between repeat measurements of radon gas taken in the same dwelling on different occasions in the same areas as the study and under approximately the same conditions as in the study were sought from the laboratories that had performed the radon measurements for the 13 studies (table 30). Data were not available for all the countries in which the studies had been carried out. Nevertheless, they provide strong evidence that there is an appreciable degree of variability between repeated measurements in the same dwelling, and they give an approximate estimate of its size.
Analyses of data from Germany (41), Sweden (R Falk, personal communication), and the United Kingdom (38) have all shown that there is greater variability between repeated measurements in dwellings with high radon concentrations than in dwellings with low radon concentrations, but that, after logarithmic transformation, the variability is approximately independent of the radon concentration. Therefore, estimates of variability were made after logarithmic transformation, and, in the statistical model used for the analysis, it was assumed that the variability was multiplicative rather than additive. The size of the variability of the repeated measurements differed from one dataset to the other, and, in particular, a much lower variability was observed in the Italian dataset than elsewhere (table 30). The Italian data are likely to be highly representative of the random measurement uncertainties in the Italy case-control study, because they were obtained in a sample of dwellings from the same area as the case-control study, and using the same radon measurement technique and protocol, the same measurement laboratory and personnel, and the same detector batch. The other datasets were also likely to be reasonably representative of the random uncertainties of the case-control studies in the countries in which they were carried out. One possible difference between the Italian dataset and the other datasets included in table 30 is that both the building material and the subsoil under the building may have contributed to the residential radon concentration in the region where the Italy study was conducted (85), whereas elsewhere the primary source was predominantly the subsoil. This difference should be present also in the corresponding case-control studies.
Although the estimates of the variability of repeat measurements were subject to considerable uncertainty, the estimated increase in the relative risk per 100 Bq/m 3 was relatively insensitive to the precise values used, decreasing only from 0.16 to 0.14 when the estimates of variability were decreased by 30% and increasing only to 0.19 when they were increased by 30% (table 31). The value of the corrected estimate obtained using the regression calibration method was virtually identical to those obtained using the more rigorous integrated likelihood method.
As was expected, correction for random uncertainties in the assessment of radon concentrations increased the residential radon concentrations among the persons with low observed concentrations and decreased them among those with high observed concentrations. As a consequence of the assumption that the uncertainties were multiplicative, the amount by which the high radon concentrations were reduced was much greater than the amount by which the low radon concentrations were increased (table 33). However, the dose-response relationship remained approximately linear after correction for uncertainties (figure 7). The effect of making separate corrections for current cigarette smokers, ex-smokers, and lifelong nonsmokers is discussed in the section Radon and Smoking, on page 49. When separate corrections were made for small-cell and other microscopically confirmed lung cancers, the estimated increase in relative risk per 100 Bq/m 3 rose to 0.51 (95% CI 0.18-1.09) for small-cell lung cancers, and the dose-response relationship remained approximately linear (figure 9), while the estimated risk increase remained much smaller and not significantly different from zero for other microscopically confirmed tumors.
The preceding corrections take into account the random uncertainties present in the assessment of residential concentrations in the case-control studies but, as the assessment of the residential radon concentrations was based on measurements of radon gas made in the recent past, it was not possible to take into account systematic factors such as the tendency for radon concentrations to have increased by around 30% between the 1950s and the 1990s, as is thought to have taken place in Sweden (86) and may also have occurred in some other countries, particularly those experiencing severe winters. It was not possible to take this factor into account in the present analysis, as there was too little information available on the amount by which the radon concentrations were likely to have increased in individual dwellings. However, if it had been possible to take this factor into account, then it would have tended to increase the estimated excess relative risk per unit radon concentration. Recent research has led to the development of methods to estimate the historical average radon concentrations to which people have been exposed, through measurements of the surface activity on a glass object that has been in a person's dwellings during the entire period of interest (87)(88)(89). Several epidemiologic studies using this technology are currently underway, and, to date, the results of two have been published (90,91). Although the uncertainties associated with this method are not yet fully understood (92,93), both of these studies have provided larger estimates of the excess relative risk associated with residential radon than corresponding estimates using recent measurements of radon gas concentrations.
An additional issue is the fact that, in the Collaborative Analysis, risk was considered in relation to the residential radon concentration rather than in relation to the person's bronchial dose from residential radon. For a fixed residential radon concentration, defined as a weighted average of living room and bedroom values, as in the Collaborative Analysis, individual persons experience a range of bronchial doses depending on the variation in the radon concentrations within the home, the amount of time the person spends in the different parts of the home, the amount of time the person spends away from the home, and the radon concentrations in the places outside the home where time is spent. If taken into account, it seems likely that this variation would also tend to increase the estimated excess relative risk per unit radon concentration (94).

Radon and smoking
As is well known, the risk of lung cancer from smoking, and especially from smoking cigarettes, is substantial. In the present study, there was negative confounding between residential radon and smoking. The estimated increase in the relative risk of lung cancer per 100 Bq/m 3 was 0.023 when study, region, age, and sex, but not smoking, were included in the stratification. This value increased to 0.052 after stratification for smoking in seven categories (never-smokers, current cigarette smokers of <15, 15-24, and ≥25 cigarettes per day, exsmokers of <10 years and ≥10 years' duration, and others), and increased further, to 0.084, when current smokers were further stratified by age at starting to smoke and ex-smokers were stratified by amount smoked (table 9).
When the analysis based on the observed radon concentrations was repeated separately for broad categories of smokers, the estimated excess relative risk of lung cancer per 100 Bq/m 3 observed radon concentration was slightly higher for the ex-smokers than for the current smokers (excess RR 0.082, 95% CI 0.003-0.211, versus excess RR 0.070, 95% CI -0.014-0.217) (table 18), and higher still for lifelong nonsmokers (excess RR 0.106, 95% CI 0.003-0.280). When corrections were made for the effects of random uncertainties in the assessment of radon concentrations, the estimated effect of radon increased in all three smoking groups, and the corrected values were 0.10 (95% CI -0.06-0.38) for current smokers, 0.22 (95% CI 0.02-0.57) for ex-smokers, and 0.20 (95% CI 0.02-0.52) for lifelong nonsmokers (table 34). Thus the size of the correction for uncertainties was greater for the ex-smokers and lifelong nonsmokers than for the current cigarette smokers. This result was due to the fact that the variability between the repeated measurements of radon gas differed between the different studies, as did the distribution of smoking histories (table B7 in appendix B). Although the estimates of the effect of radon on the relative risk of lung cancer differed appreciably between the three main smoking categories, the confidence intervals were so wide that there was no formal evidence of heterogeneity between the categories, and the data are compatible with the relative risks when the same value is used for all three smoking categories.
The presence of random uncertainties in the assessment of residential radon concentrations means that risk estimates based on measured radon concentrations inevitably underestimate the true magnitude of the risks.
In the present study it was possible to make an approximate adjustment for the effects of random uncertainties, and this adjustment increased the estimated effects of residential radon substantially. It seems certain that there were also random errors in the assessment of smoking habits. As there was negative confounding between smoking and radon in these data, a correction adjusting for such uncertainties would be likely to increase the estimated effect of exposure to radon still further, particularly among current smokers, and, in principle, it would be desirable to make such an adjustment (41). However, there is little quantitative information available on the nature and extent of random errors in the reporting of smoking histories in the context of studies such as this. In addition, making such adjustments in the context of a complex analysis would present considerable technical challenges. Therefore such adjustments were not attempted, nor were such adjustments made in any of the other major studies of the joint effects of radiation and smoking.
The effect of radon exposure among persons with different smoking histories has also been examined among miners occupationally exposed to radon (2,66). For 7 of the 11 studies for which data were available, there was some, although usually very limited, information on smoking status. When those reported to be lifelong nonsmokers (among whom there were 64 deaths from lung cancer) and others were considered separately, the estimated excess relative risk per unit radon concentration was 3.03 times higher than the corresponding ratio for those who were nonsmokers. Despite the large size of this ratio, these estimates had wide confidence intervals, and the difference between them was not statistically significant.
The effect of exposure to ionizing radiation on persons with different smoking histories has recently been examined in the life-span study of the survivors of the atomic bombings of Hiroshima and Nagasaki (95), where the exposures were predominantly to gamma radiation. In this population there was strong evidence that the excess relative risk per sievert varied within the categories of smoking level (P<0.01). There was a tendency for the risks relative to those of unexposed persons with the same smoking level to decrease with an increasing amount smoked, and, in the age group of 60-70 years, the estimated excess relative risk per sievert was 0.64 [standard error (SE) 0.45] for the nonsmokers and 1.02 (SE 0.66), 0.10 (SE 0.14), and 0.00 (SE 0.14) for the smokers of 1-15, 16-25 and ≥26 cigarettes per day, respectively. The ex-smokers were omitted from this analysis. When radiation exposure and smoking were considered simultaneously for the survivors of the atomic bombings, a multiplicative model was rejected, but the data were consistent with an additive model (ie, one in which the increase in the absolute risk of lung cancer per sievert was the same, regardless of smoking history).
In all of the three studies referred to in the previous paragraph, the excess relative risk per unit exposure was lower for continuing smokers than for nonsmokers, although, in the present dataset and in the studies of radon-exposed miners, the difference did not reach statistical significance. The consistency of the finding may suggest that the effect of radon is, in fact, proportionately somewhat greater for nonsmokers than for smokers. In the present study, however, there was negative confounding with smoking and, although it had been taken into account as far as is possible, it seems likely that some residual confounding with smoking remained, particularly for current smokers, due to random errors in the reporting of smoking histories. Thus, in the present study, the observed differences between continuing smokers and lifelong nonsmokers in the excess relative risks per unit radon concentration may also be simply a reflection of an inadequate allowance for confounding. Therefore, in the analyses of the joint effects of smoking and radon presented in this publication, we have assumed that the excess relative risk per 100 Bq/m 3 is the same, regardless of smoking status.
When the implications of the risks of exposure to residential radon, especially in the context of public health, are considered, it is desirable to consider jointly the effects of radon and the likely risks of lung cancer from smoking. In the present analysis, the relative risks of current and ex-smokers relative to that of lifelong nonsmokers (table 3) were similar to those found in other studies carried out with these populations (31). In the present analysis the joint effect of smoking and radon has been estimated by assuming that the relative risk for persons in each broad smoking category was known precisely and was equal to the relative risk that was found for all men in that smoking category when they were compared with lifelong nonsmokers (tables 37 and 38, and figures 10 and 11). [A discussion of why this was done is given under "Joint Effect of Smoking History and Radon Exposure on Lung Cancer Risk" in the Statistical Methods section.] For the persons in the collaborative analysis, the mean time-weighted corrected radon concentration during the period of interest was 90 Bq/m 3 for the lung cancer cases and 86 Bq/m 3 for the controls (table 32) and only 7.7% of them had TWA corrected radon concentrations of >200 Bq/m 3 . Thus, for the vast majority of the persons included in this analysis, the risk of lung cancer was determined predominantly by their smoking history rather than by their radon exposure (figures 10 and 11). The component studies of the Collaborative Analysis were, on the whole, carried out in high radon areas of the countries concerned, and several studies included only persons with low residential mobility.
Summaries of the results of the surveys of the indoor radon concentrations that have been carried out for the general population in different countries are given in the reports of the United Nations Scientific Committee on the Effects of Atomic Radiation (1). According to these surveys, the population-weighted average indoor radon concentration for Europe as a whole is around 60 Bq/m 3 , while the worldwide average is 39 Bq/m 3 . Thus with respect to the general population, both in Europe and worldwide, most people have residential radon concentrations that are <200 Bq/m 3 . At these concentrations, for those who smoke cigarettes, a far greater reduction in the cumulative risk of lung cancer would likely be attained if they gave up smoking than if they reduced the level of their radon exposure.

Quantitative comparison of the risks observed in the present study with those observed for underground miners occupationally exposed to radon
In a collaborative analysis of 11 cohorts of underground miners of uranium and other igneous rocks who were occupationally exposed to radon, there were substantial inverse dose-rate effects (2,66). This finding may be due, at least in part, to the effects of random errors in the assessment of radon exposures among the miners, although several other mechanisms have been postulated (2,96). In the Collaborative Analysis of the 11 miner cohorts, the average exposure received by the miners was 158 working level months (WLM), 19 received over an average of 5.7 years (2,66). In contrast, the mean residential radon concentration experienced by the populations represented in the studies included in the Collaborative Analysis was 86 Bq/m 3 after correction for uncertainties in the assessment of the radon concentration. Living in a home with a radon concentration at this level for 30 years results in an exposure of approximately 10 WLM, 20 less than one tenth the average cumulative exposure of the miners.
The generally higher exposures and also the existence of inverse dose-rate effects in the miners' data complicate the quantitative comparison of the risks of lung cancer between the present study and that of the occupationally exposed miners. The most relevant comparison is with an analysis of a subset of the data from the 11 studies of miners that was limited to total exposures of <50 WLM (2,75). The bronchial dose received by such an exposure would be approximately equal to that incurred by living in a home with a concentration of 427 Bq/m 3 for 30 years. Within this group, the estimated excess relative risk for those exposed at concentrations below 0.5 WL was 0.0251 per WLM, which is equivalent to an increase in the relative risk of 0.30 per 100 Bq/m 3 when residential exposure over the previous 30 years is taken into consideration, while removing the restriction to total exposures below 50 WLM suggested an increase of 0.19 in the relative risk per 100 Bq/m 3 . In a more recent follow-up of one of the miner cohorts in which the average cumulative exposure was only 36.5 WLM, the estimated excess relative risk per WLM for those exposed only to concentrations lower than 1 WL was similar to that for the miners with total exposures of less than 50 WLM, at 0.024 per WLM (97).
The increase in the relative risk per unit exposure seen in the collaborative analysis of data from the 11 miner cohorts with total exposures below 50 WLM is compatible with, although somewhat higher than, the estimated value of 0.16 (95% CI 0.05-0.31) per 100 Bq/m 3 after correction for uncertainties in the assessment of the radon concentrations that was obtained in the present Collaborative Analysis. When the two results are compared, the uncertainties in them both should be borne in mind. The uncertainties in the estimate from the present Collaborative Analysis have already been discussed. However, there are also substantial uncertainties for the estimated excess relative risk from the miners' studies. Confidence intervals have not been published for the estimates based on the 11 cohorts of miners cited in the previous paragraph. In addition, there was substantial heterogeneity between the effects of radon in the different 19 The working level (WL) is defined as any combination of the short-lived radon progeny in one liter of air that results in the ultimate release of 1.3 x 10 5 MeV of potential α-particle energy. Exposure to this concentration for a "working month" of 170 hours (or twice this concentration for half as long, and so forth) is defined as a working level month (WLM). 20 This follows if it is assumed that 1 Bq/m 3 at equilibrium is equivalent to 0.00027 WL, that the equilibrium factor (see below) in dwellings is 0.40, that people spend 70% of the time at home, that there are 365.25 × 24 / 170 = 51.6 "working months" (see footnote 19) in 1 year, and that the ratio of the dose to lung cells for exposures in homes to that for similar exposures in mines (sometimes referred to as the K-factor) is unity. Under these assumptions, living in a home with 86 Bq/m 3 for 30 years will result in 86 × 0.00027 × 0.40 × 0.70 × 51.6 × 1 × 30 = 10.1 WLM. (The equilibrium factor is defined as the ratio of the equilibrium equivalent concentration of radon to the actual radon concentration, for which the equilibrium equivalent concentration is the activity concentration of radon in equilibrium with its short-lived progeny that has the same potential alpha energy concentration as the actual nonequilibrium mixture.) Scand  miners' studies (2,66). It should also be borne in mind that the effect of uncertainties in the assessment of radon exposures has not been taken into account in the main analyses of the miners' studies. In the study of a single cohort of miners in which it was taken into account, it was found that the estimated excess relative risk was increased by around 60% for high dose-rate exposures, but was little changed for low dose-rate (0-15 WL) exposures (98).

Concluding remarks
The Collaborative Analysis of data from 13 European studies of the effects of residential radon on the risk of lung cancer presents, for the first time, strong evidence of an association between residential radon and the risk of lung cancer after adjustment for smoking history. The dose-response relationship was linear with no evidence of a threshold; it remained statistically significant even when the analysis was limited to observed radon concentrations of <200 Bq/m 3 ; and there was no evidence that the excess relative risk per unit increase in the radon concentration varied with smoking history. When based on the observed radon concentration, the estimated excess relative risk per 100 Bq/m 3 was 0.084 (95% CI 0.03-0.158). However, this estimate is likely to underestimate the true risk, as it does not take into account the uncertainties present in the assessment of residential radon concentrations. When an approximate correction for the effect of such uncertainties was included, the estimated excess relative risk per 100 Bq/m 3 was 0.16 (95% CI 0.05-0.31). These results are crucial to the development and refinement of policies for managing exposure to this form of natural radiation, so as to help reduce the annual number of deaths from the most common type of fatal cancer in Europe. In addition, in every study there were substantial numbers of research assistants, and other staff whose contributions to the data collection and collation and the conduct of the study were essential, and the authors gratefully acknowledge their contributions. The authors would also like to thank the persons who took part in the study and the assistance provided by the local medical and other staff in each study area. Finally, the authors would also like to express their thanks to David Cox, Tom Fearn, Jon Miles, and Richard Peto for their helpful discussions during the preparation of this report.

Acknowledgments
The Collaborative Analysis was funded by Cancer Research UK, the European Commission [Nuclear Fission and Radiation Protection Programme contracts FIGH-CT1999-00008 and 516483 (FI6R)], and also the authors' institutions. The funding sources had no role in the study design, the data collection, the data analysis, the data interpretation, the writing of the report, or the decision to submit the paper for publication. persons who had died from smoking-related causes) in the same period taken from the same source, individually matched by sex, age, and year of the decedents. Only German-speaking people for whom a next-of-kin could be found were selected for the study.
There was no review of pathological material. The index year was the year of death.

Czech Republic
The study was based on a cohort study of the population of the Middle Bohemian Pluton.  (3)] were identified from the Finnish Cancer Register. There was no specific review of pathological material. One control matched by year of birth and sex was identified from the same cohort for each case, the control had to be alive at the time of the diagnosis for the corresponding case. In the event that no information was available for the primary control another control was identified.The index year was the year of diagnosis for the cases. For the controls it was the year of diagnosis of the matched case or, if this information was not available, the mean year of diagnosis for the cases with the same sex and year of birth. Finland southern Nineteen rural municipalities in southern Finland with high radon levels were included in the study. All men with lung cancer diagnos-[Ruosteenoja et ed in the resident population in these municipalities during 1980-1985 were identified through the Finnish Cancer Register (1980al, 1996 (4)] 1982) or through hospitals in the study area with diagnostic and treatment facilities for lung cancer (1983)(1984)(1985). The cases were excluded from the study if they were institutionalized, had migrated to the study area after 1 January 1980, had been diagnosed prior to 1980, had a lung metastasis only secondary to another primary site, or had missing case records. There was no specific review of pathological material. A random sample of men resident in the study municipalities on 1 January 1980 and balanced according to the age distribution of the cases was drawn from the National Population Register. The controls were excluded from the study if they were institutionalized, deceased with no next of kin, or had been diagnosed with lung cancer. A questionnaire on smoking was sent to the remainder. Among those who replied, all current smokers were selected for the study, as were random samples of approximately 10% each of ex-smokers and lifelong nonsmokers.The index date was the year of diagnosis for the cases and the mean year of diagnosis of the cases (1982) for the controls. France [Baysson The study was undertaken in five radon-prone regions (Auvergne, Brittany, Languedoc, Limousin and Ardennes). In each participating 2004 et al, (5)] hospital where the investigation and treatment of lung cancer is carried out, all persons diagnosed with primary lung cancer during 1992-1998 and under 75 years of age on their date of diagnosis were identified by local interviewers and interviewed within 6 months of diagnosis. Persons were eligible for the study if they had lived in the study areas for at least 25 out of the last 35 years. There were no exclusion criteria based on occupation or nationality. There was no specific review of pathological material. For each case, two controls were identified from the pool of hospital patients of the same region who satisfied the study residence requirements and who had not been hospitalized for a disease strongly related to tobacco. Within each region the controls were individually matched to the cases for sex and age (± 5 years). In each region, attempts were made to include patients with a wide variety of diseases in the hospital control group. The index year was the year of diagnosis for both the cases and the controls. Germany eastern The cases were recruited in five hospitals in Thuringia and Saxony from October 1990 to March 1997 and were eligible if they current-[Wichmann et al, ly resided in the study area, had lived in Germany since 1965, were aged 75 years or less, had microscopically confirmed lung cancer 1999 (6); Kreuzer with no evidence of tuberculosis, could be interviewed within 3 months of the initial diagnosis, had never worked in the SDAG (Die et al, 2003 (7); Sowjetisch-Deutsche Aktiengesellschaft) Wismut uranium mining industry, and were well enough to be interviewed personally. Wichmann et al, Histological material was reviewed by one pathologist and cytological material by another. Histological and/or cytological material was 2005 (8)] available for all the cases; a review of the histological material by one reference pathologist and of the cytolgical material by another was performed for 61% of the cases. Population controls were recruited during 1990-1997 using a two-stage process. First, a random sample of about 30 000 people was selected from the central registration office of the former German Democratic Republic. Then people were selected at random with frequency matching to the cases for sex, age (six 5-year groups), and 10 matching regions. The eligibility criteria were similar to those for the cases.The index year was the year of interview (for the cases within 3 months of diagnosis). Exclusion criteria were as for the Germany western study. Germany The study was carried out in the rural areas of Eifel, Westerwald/Hunsrueck, and Upper Palatinate/Lower Bavaria and also in neighwestern [Wich-boring urban areas (the industrialized Ruhr area, Rhineland and the Saar region). The cases were recruited in 9 hospitals from Octomann et al, 2005 ber 1990 to October 1995 and were eligible if they currently resided in the study area, had lived in Germany since 1965, spoke (8); Wichmann German, were aged 75 years or less, had microscopically confirmed lung cancer with no evidence of tuberculosis, could be interet al, 1998 (9); viewed within 3 months of the initial diagnosis, and had never worked in the SDAG (Die Sowjetisch-Deutsche Aktiengesellschaft) Kreienbrock Wismut uranium mining industry. Histological and/or cytological material was available for all of the cases; a review of the histoloet al,2001 (10)] gical material by one reference pathologist and of the cytological material by another was performed for 75% of the cases. People were excluded when the current address had been occupied for less than a year or when the subject had spent less than 25% of his or her time at home. Population controls were selected at random from registers of residents or by random digit dialing, were frequency matched to the cases for sex, age (six 5-year groups), and 13 matching regions. The eligibility criteria were similar to those for the cases. The index year was the year of interview (for the cases within 3 months of diagnosis).
(continued)  The study was carried out in the Santiago de Compostela Public Health District in northwest Spain and was population-based. All of Dios et al, 2002 the cases of lung cancer diagnosed in the period 1992-1994 were identified from the pathology departments of the local hospitals. (12)] There was no specific review of pathological material for this study. The controls were selected at random from the 1991 census, subject to frequency matching to the cases by geographic area (comarca) of current residence and sex. Exclusion criteria were age under 35 years, current residence less than 5 years' duration or having undergone major structural alterations recently, cases without histological confirmation and/or undergoing occupational or therapeutic radiation, controls with respiratory tract disease or previous cancer. The interviews took place during the period 1992-1995 and were carried out by two specially trained interviewers, either with the subject in person or, if he or she had died, with their closest surviving relative or cohabitant. The index date was the mean year of diagnosis of the cases (1993) for all the participants. Sweden nation-The study was carried out in 109 municipalities in Sweden, of which 56 had been selected as high-radon areas and the remainder as wide [Pershagen low-radon areas. The base population for the study included all persons 35 to 74 years of age who had lived in one of these municet al, 1994 (13) (17)]: never-smokers were selected from a case-control study of air pollution and lung cancer. The cases were selected from those diagnosed between 1 January 1985 and 31 December 1990 in the Stockholm County regional cancer registry and were included if they were aged 40-75 years, resident in Stockholm county at some time between 1 January 1985 and 31 December 1990, with no more than 5 years of residence outside the county between 1950 and 1990. About twice as many controls were randomly selected in 1992 from population registers of Stockholm county for the corresponding years, frequency-matched for age at calendar-year of diagnosis. About half of the controls were additionally frequency-matched to the cases for vital status in 1990, excluding those with a smoking-related cause of death. Study 3 [Pershagen et al, 1995 (18)]: a study designed to assess lung cancer risk from environmental emissions from a large smelting plant in northern Sweden recruited cases and controls from men and women who died in the Skellefteå Municipality between 1961 and 1990, excluding those who worked at the plant or as miners. Two controls per case were selected, matched for birth year and sex. For the present study 40 never-smoking cases, 78 never-smoking controls who had died in the period 1980-1990 were selected. Study 4 [Axelsson et al, 1996 (19)]: a case-control study of environmental factors and lung cancer was conducted among people aged less than 75 years and living at the time of the study in one of 26 municipalities in Göteborg and Bohus county or Ålvsborg county in southwest Sweden. The cases were recruited from the three local regional hospitals from January 1989 to June 1994. The controls were selected from population registers matching for county, sex and birth date. All 49 never-smoker cases and twice as many never-smoker controls were selected for the present study. Study 5: A case-control study investigated the association between occupational exposure and lung cancer in the county of Västernorrland, where many paper and pulp mills are situated. The cases were men diagnosed with lung cancer between 1978 and 1991, as recorded in the Umeå Regional Cancer Register, who died before 1 September 1992. There was no pathological review specifically for the present study. With the controls for each study one control group was selected from the National Cause-of-death Register (excluding deaths from suicide or lung cancer) and matched to cases on year of birth, year of death and municipality of residence at the time of diagnosis of the matched case. A second control group was selected from persons alive at the time of diagnosis of the case using the national population register, matching for year of birth and municipality of residence. All 25 never-smoker cases diagnosed in 1980 or later and twice as many controls were included in the study. For all these five component studies the index year was the year of diagnosis for the cases. For the controls it was the year of diagnosis of the case for living controls who died less than 3 years before their matched case and the year of death for the controls who died more than 3 years before the year of diagnosis of the matched case. Sweden Stock-Females referred with suspected lung cancer were interviewed at all the clinical departments of pulmonary medicine and thoracic holm [Svensson surgery at hospitals in Stockholm county from September 1983 through December 1985. The final diagnosis was obtained from et al, 1989 (20); hospital records. Those subsequently found not to have lung cancer constituted the hospital group. There was no specific review of Pershagen et al, pathological material for this study. For each case of lung cancer a population control was randomly selected from Stockholm 1992 (21)] County population registers from women who were born on the same day as the case. The index year was the year of interview for both the cases and controls. United Kingdom The study was carried out in Devon and Cornwall, the two counties in Great Britain with the highest average residential radon [Darby et al, concentrations. People under 75 years of age who were referred with suspected lung cancer during a 4-year period were identified 1998 (22)] at five hospitals in Devon and Cornwall, where lung cancers are investigated and treated. They were eligible for the study if they were current residents of Devon or Cornwall and had lived in either county for at least 20 years during the 30-year period ending 5 years previously. Only those who were ethnically white were included in the study. Pathological specimens were reviewed by one pathologist. For each person with suspected lung cancer, a control was sought from hospital patients of the same sex, born within 5 years of the case, who satisfied the study residence requirements, whose current hospital admission was for a disease not known to be strongly associated with smoking, and who lived in the same area as the case. For each hospital two or three areas were defined, one being the local area of the hospital and the other(s) more distant areas. A population-based group of controls was selected and was frequency-matched to the persons with suspected lung cancer by age, sex and county of current residence.