Residential Radon and Lung Cancer: The Canadian Connection
Selon Santé Canada, le radon est la deuxième cause principale de cancer des poumons. Qu’ils travaillent directement avec le radon ou pas, les spécialistes de la radioprotection se font souvent questionner au sujet de cet élément chimique par leurs familles et amis. Toutefois, Robyn Reist se demande combien de spécialistes de la radioprotection sont suffisamment informés pour faire plus que simplement régurgiter l’information fournie par Santé Canada. Dans son article, elle livre une évaluation en profondeur de cette recherche d’envergure pour en résumer les grandes lignes. Elle espère ainsi doter les lecteurs de meilleures réponses au sujet du radon, et les enjoindre à discuter sur le sujet.
Robyn Reist, University of Saskatchewan
MSc student, Department of Community Health and Epidemiology
According to Health Canada, radon is the second-leading cause of lung cancer—16% of lung cancer deaths in Canada are attributed to radon. Radon exposure is quite literally the radiation protection issue that is closest to home for the majority of Canadians. Not surprisingly, those of us in the radiation protection community, whether we work directly with radon or not, are often asked about this topic by our families and friends.
How many radiation protection specialists, however, are informed enough to do more than regurgitate information provided by Health Canada?
Using the only Canadian epidemiological study on residential radon and lung cancer as a starting point, I will summarize the major research and provide in-depth assessment. I hope to equip you with better answers about radon and inspire you to participate in discussions on the topic
Attempts to measure radon concentrations in homes
Let’s begin with Létourneau’s 1994 article, “Case-Control Study of Residential Radon and Lung Cancer in Winnipeg, Manitoba, Canada.” I want to examine how this single study has contributed to the wider scientific body of literature and influenced public policy.
Létourneau’s article describes the only Canadian case-control study of residential radon and lung cancer conducted to date. It was one of the first epidemiological studies in the world to attempt to measure radon concentrations in current and previous homes of its subjects.
While the results were inconclusive, the study contributed to a wider body of literature on the topic and was included in a large North American meta-analysis that influenced the Canadian residential radon limit change from 800 Bq/m3 to 200 Bq/m3 in 2007.
The Winnipeg study
The study reported by Létourneau et al. has many strong points that highlight the efforts researchers must undertake to collect meaningful data in this area. It also has several limitations and sources of bias that potentially make its results inconclusive.
To briefly summarize, study subjects were recruited from the Winnipeg population between 1983 and 1990. Subjects with lung cancer (cases) were recruited from the Manitoba provincial cancer registry. Healthy subjects (controls) were matched with the cancer subjects. Age, sex, smoking status, and education were all controlled for, either via the study design or statistical analysis. In total, there were 738 case-control pairs (1,476 subjects).
For one year, radon was measured using alpha-track detectors in as many Winnipeg residences that the subjects had lived in as possible. Radon concentrations were very similar in the homes of cases and controls, although slightly higher in the homes of control subjects. The results were inconclusive and did not show a statistically significant relationship between lung cancer risk and residential radon exposure.
The Swedish studies
The Létourneau study design seems to be based on two Swedish studies led by Göran Pershagen. In two different case-control studies that monitored residential radon, Pershagen found odds ratios greater than 1.5 for radon concentrations above 140 Bq/m3. However, one of the studies had roughly triple the sample size of the Winnipeg study, and the other selected nearly half of its control subjects from a hospital setting, which may have introduced bias.
Differences in sample size and subject population from the Swedish studies may have created a situation where using traditional power calculations to dictate sample size and obtain a similar odds ratio might not be adequate. It is interesting to note that Pershagen provided his own critique of the Winnipeg study as a letter to the editor in the American Journal of Epidemiology in 1995 and raised some valid concerns.
Analysis of the Winnipeg study
Case-control observational design
The case-control observational design using incident cases used in the Winnipeg study improved on previous North American ecological and case-control studies and attempted to add some consistency to the literature at the time (i.e., the Swedish studies). Attempting to monitor residential radon in all of the homes of the study subjects was a large improvement over previous North American case-control studies that had monitored radon in only one home.
The Winnipeg study was conducted around the same time as a number of similarly designed studies in North America and Europe. This is likely not a coincidence—large meta-analyses using the results of many of these studies were completed approximately 10 years later. Using the pooled data, studies by Krewski et al. and Darby et al. found statistically significant (though just barely) odds ratios showing a dose-response relationship between lung cancer and residential radon exposure.
There were a few significant discrepancies between the case and control populations in the Winnipeg study.
There were a few significant discrepancies between the case and control populations in the Winnipeg study. After lung cancer cases were confirmed and subjects enrolled in the study, control subjects were chosen at random from the telephone book. While this was a valid way to randomly select subjects in the 1980s, no further details were provided as to how selection bias was mitigated during enrolment of controls.
For example, even the time of day the phone calls were placed could have biased the control group towards a certain socioeconomic status. If the calls were placed during the day, any men who answered were more likely to be shift workers or unemployed (possibly of lower socioeconomic status). Women who were home during the day likely did not work because their husbands earned enough to support the family (higher socioeconomic status).
Case-control pairs were matched by age and sex in order to perform a paired statistical analysis. While cases and controls were indeed from the same source population of Winnipeg (meaning controls would have become cases if they had developed lung cancer and cases could have been controls had they not developed lung cancer), there may still have been important differences between the two groups that could impact the odds ratios.
In Létourneau’s article, there is no discussion of how long the subjects actually lived in Winnipeg as a percentage of their lifetime, and it is unknown whether this was considerably different between cases and controls.
The controls were found to:
- smoke less (70% of controls were current or former smokers vs. almost 97% of cases)
- be better educated (83% of controls had finished secondary school vs. 70% of cases)
- be more likely to have been born in Canada
These differences could have also influenced how much time people spent in their homes. For example, women of higher socioeconomic status might spend more time in their homes, whereas people of lower socioeconomic status in general spend less time at home because they work multiple jobs or shift work. Alternatively, if someone is unemployed, they likely spend a lot of time at home.
Given the differences in education/socioeconomic status, it seems likely that case subjects and control subjects would have spent different amounts of time in their homes. There are a lot of potential variables relating to socioeconomic status, and the fact that researchers didn’t consider them introduces considerable potential bias into the study.
Measuring radon in residences
Demographic information and residences were identified via telephone interviews with the study subjects, or proxy interviews in the cases where subjects were deceased. Subjects were not told what the study was investigating, and radon was not mentioned until the subjects were asked if their current home could be monitored.
There are potential issues with the measurement of the exposure to residential radon, but any errors would have likely been applied consistently to both cases and controls.
All residences in the Winnipeg area in which subjects had lived over the past 30 years were identified. Researchers did not expect to be able to monitor all residences for logistical reasons, but they were able to monitor three different residences, on average, for each subject.
An attempt was made to measure radon concentrations in the rooms where the lowest radon concentration would be expected (the bedrooms) and the highest concentration would be expected (the basement). These measurements represented “best- and worst-case scenarios.”
One of the strengths of the Winnipeg study was the attempt to look at the exposure in this manner.
One of the strengths of the Winnipeg study was the attempt to look at the exposure in this manner. Averaging basement and bedroom values would not have been appropriate due to the disproportionate amount of time most people spend in their basements.
The same measurement and analysis protocols were applied to all available residences, and this seems to have been undertaken by a single lab to ensure consistency. Alpha-track monitoring was not highly accurate at the time of the study (there was a known error of +/− 25%), but efforts were made to ensure consistency even with a large known error. They also used two dosimeters in each room and took an average. (I would have used three to ensure that any spurious results could be more easily identified.)
Table 1: Radon concentrations in the homes of subjects in the Winnipeg study
|Case subjects||115.5 Bq/m3||188.5 Bq/m3|
|Control subjects||125.6 Bq/m3||206.0 Bq/m3|
Overall, the concentrations of radon measured in the homes of cases and controls were extremely similar, with a slightly higher concentration found in the homes of controls. This could potentially be explained by the fact that the controls were generally of higher socioeconomic status and thus may have lived in newer and better-insulated houses, which limit airflow and create more opportunity for radon levels to build up in a dwelling.
The authors of the report did not consider whether renovations or other changes to buildings might have altered the radon concentrations since the subjects had lived there. The differing socioeconomic status between cases and controls meant there was a possibility that this might have occurred more on one side than the other.
The researchers performed a variety of matched conditional logistic regression models to test for a relationship between residential radon and lung cancer, but they were consistently unsuccessful.
Paired analysis is a valid way to control for confounders via matching. In this study, age and sex are the matching variables, so, for example, a 45-year-old male case is matched to a male control of approximately the same age (+/− 5 years in this study).
The final statistical model included smoking and education as predictors, but country of birth and occupational exposures were not found to significantly affect the results, so they were not included in the final model to calculate the odds ratios.
In this study, only a very small number of subjects were exposed to occupational lung carcinogens, which is likely why this variable was not found to be statistically significant. A significance test tests only for predictive, not confounding, effects of a variable. Using this method to include or exclude variables can lead to inaccurate modelling or exclusion of important confounders.
Effect modification not considered
The report did not mention effect modification. Radon has been established to have an interactive effect with smoking, and, in the introduction, the authors stated that other studies had observed a “synergistic effect” between smoking and radon exposure. However, there is no description in the paper of whether smoking was tested as an effect modifier (or even as a confounder, as described above).
I thought this oversight was strange, so I looked to the two larger meta-analyses of similarly designed studies (mentioned earlier). Neither of these articles found evidence of smoking interacting with residential radon exposure in the pooled data.[17,18] Perhaps at the low levels measured in these studies, effect modification could not be confirmed. Alternatively, the differences in smoking classification between studies may have made this too difficult to perform in the meta-analyses.
Potential issue with imputing data
Missing data was imputed (an assumed value was assigned) in order to ensure all data fields were populated for the paired analysis. For example, if a house did not have a basement, the researchers still applied an estimated basement measurement (1.33 × bedroom measurement) for that subject based on the existing data from other houses in the study. If there were subjects where no readings could be taken in any residences, exposures were estimated similarly. While this was a statistical requirement for the paired analysis, there are a few issues with it.
The authors do not state whether they imputed more for cases or controls, or if they were approximately equal. If either the cases or controls were more heavily weighted with imputed data (“study average” values), this could bias the results.
In his critique, Pershagen notes that imputing data based only on measurements taken in Winnipeg residences could create a biased exposure estimate, as Winnipeg had higher average levels of radon than the rest of the country. Therefore, any imputed data from years not lived in Winnipeg would likely be overestimated.
Pack years not an accurate way to account for smoking
Pershagen also raises an issue with the way smoking was classified and controlled for. He states that using pack years is not an accurate way to account for former smokers, and that using this method to analyze and control for smoking status may introduce additional confounding towards the null.
Urbanization not accounted for
He also notes that urbanization is a strong confounder for lung cancer, but how long the subjects had lived in Winnipeg was not accounted for. This second point is another potential issue created by imputing data.
Sensitivity analysis to check for bias
Sub-analyses (more commonly known today as sensitivity analyses) were performed to check for potentially obvious biases, which was one of the strengths of this study.
When the researchers performed the sub-analyses, they removed the case-control pairs that involved proxy interviews and included only case-control pairs where at least 75% and 90% of residences were able to be measured (thus reducing the amount of imputed data, but not eliminating it).
They also divided the results into two residence groups—those measured 5–30 years before the study and those measured 5–15 years before the study. A higher number of residences were measured in the 5–15 year period, and the authors note that this is likely the most relevant period to measure when considering radon-induced lung cancer. However, none of the sub-analyses resulted in a noticeable dose-response relationship or positive association.
In a comparison of all data vs. the 75% true-measurement sub-analysis, the calculated odds ratios are noticeably different between these two states—they are generally higher for the “all data” category but do not follow any kind of recognizable pattern. Thus it is possible that the previously mentioned potential biases related to imputing data (overestimating exposure due to applying an estimate from Winnipeg results) did have an effect on the results that biased the odds ratios in a slightly positive direction.
No link found between residential radon and lung cancer
No significant association was observed between residential radon and lung cancer in this study, and the results did not suggest a dose-response relationship.
Because no association was found, there is no chance that a type I error (false positive association) was made. There is, however, a chance that a type II error (false negative association) was made, and the authors mention this in the discussion.
As mentioned earlier in this article, the Winnipeg study was designed to look for a similar association as was found in the Swedish studies by Pershagen et al. (having an 80% chance of detecting an increased odds ratio greater than 1.5). Power calculations based on these parameters defined the study size, which was potentially too small when considering all of the other potential sources of bias and error that needed to be taken into consideration.
While several other studies conducted at the time also found no association, the Winnipeg researchers were obviously hoping to find results consistent with the Swedish studies. They did not.
Approximately a decade later, when the data from the Winnipeg study was pooled with other similar North American studies, the new results were consistent with both the 2004 European meta-analysis by Darby et al. and an extrapolation of uranium-miner data performed to estimate risk from residential radon. It is worth noting that none of these studies was able to replicate odds ratios as high as the Swedish studies.
Canadian guidelines revised
In 2006, the Canadian Radon Working Group submitted its recommendations for a new radon guideline for Canada, namely to lower the action level from 800 Bq/m3 to 200 Bq/m3. The report provides an excellent summary of the existing body of knowledge on residential radon and lung cancer, and the working group’s reasoning is sound for lowering the limit. The majority of literature appears to be pointing in the same direction (albeit a very weak direction).
The consistency between the North American and European meta-analyses and the Committee on Health Effects of Exposure to Radon’s extrapolation was influential in providing the rationale for lowering the radon limit. However, when reading the appendix discussing residential radon exposure and lung cancer’s association via Bradford Hill’s nine criteria of causation (strength, consistency, specificity, temporality, dose-response, plausibility, coherence, experimental intervention, and analogy), one begins to realize there is still so much we do not know about this relationship.
One begins to realize there is still so much we do not know about this relationship.
One statement in particular stood out to me—the working group declares that specificity is partially satisfied because the only known effect of radon is lung cancer. However, this can be disputed—in Japan, radon has been experimentally shown to have anti-inflammatory properties, and it is used in hospital settings to medically treat conditions such as osteoarthritis. Its acceptance as an anti-inflammatory by medical professionals is also arguably prevalent in Europe. Thus I do not think specificity is as clear as it seems for radon.
Some epidemiologists might also dispute whether an excess odds ratio under 0.2 is really enough evidence to draw a definitive conclusion. On the other hand, some recent studies are suggesting more robust data-collection techniques point to the currently accepted risk being an underestimation. Regardless, with a disease as serious as lung cancer, it is absolutely better to be safe than sorry.
Even after digging into the details, I would not be personally comfortable arguing passionately for or against a definitive dose-response relationship between low-level radon and lung cancer. While I’d absolutely remediate my house if it tested high, I wouldn’t think twice about visiting a “radon spa” if I ever develop arthritis.
1. Government of Canada, 2017. Radon: Frequently asked questions.
2. Létourneau et al., 1994. Case-control study of residential radon and lung cancer in Winnipeg, Manitoba, Canada. American Journal of Epidemiology, 140(4), 310–322.
3. Krewski et al., 2005. Residential radon and risk of lung cancer. Epidemiology, 16(2), 137–145.
4. Government of Canada, 2017. Radon: Frequently asked questions.
5. An odds ratio is a measure of the association between an exposure and an outcome, or the odds that an outcome will occur given a particular exposure compared to the odds of the outcome occurring in the absence of that exposure (e.g., an odds ratio of 1.5 means the odds that people with the disease were exposed to the defined concentration of the contaminant are 50%).
6. Pershagen et al., 1994. Residential radon exposure and lung cancer in Sweden. New England Journal of Medicine, 330(3), 159–164.
7. Pershagen et al., 1992. Residential radon exposure and lung cancer in Swedish women. Health Physics, 63(2), 179–186.
8. Westreich, 2012. Berkson’s bias, selection bias, and missing data. Epidemiology, 23(1), 159–164.
9. Pershagen & Lagarde, 1995. Letters to the editor: Re: “Case-control study of residential radon and lung cancer in Winnipeg, Manitoba, Canada.” American Journal of Epidemiology, 142(10), 1121.
10. Krewski et al., 2005. Residential radon and risk of lung cancer. Epidemiology, 16(2), 137–145.
11. Darby et al., 2004. Radon in homes and risk of lung cancer: Collaborative analysis of individual data from 13 European case-control studies. British Medical Journal, 330(7485), 223–228.
12. Logistic regression calculates the likelihood of developing or not developing a condition based on the exposure to the suspected disease-causing agent. Conditional logistic regression uses matching to control for confounding (see below), so potential predictors (e.g., age, sex) do not need to be considered as predictors in the model.
13. A confounding variable is an extra variable that was not accounted for in the study design but which could influence the outcome. Confounding variables can suggest there are correlations when there really are not.
14. Sonis, 1998. A Closer Look at Confounding. Family Medicine, 30(8), 584–588.
15. Effect modification occurs when the same exposure has different effects on different subgroups (e.g., a drug is effective on female patients but not on male patients).
16. Committee on Health Risks of Exposure to Radon, 1999. Health Effects of Exposure to Radon: BEIR VI. Washington: National Academy Press.
17. Darby et al., 2004. Radon in homes and risk of lung cancer: collaborative analysis of individual data from 13 European case-control studies. British Medical Journal, 330(7485), 223–228.
18. Krewski et al., 2005. Residential radon and risk of lung cancer. Epidemiology, 16(2), 137–145.
19. Pershagen & Lagarde, 1995. Letters to the editor: Re: “Case-control study of residential radon and lung cancer in Winnipeg, Manitoba, Canada.” American Journal of Epidemiology, 142(10), 1121.
20. Pack year is a measurement of the amount a person has smoked over a period of time. It is calculated by multiplying the number of packs of cigarettes smoked per day by the number of years the person has smoked. Generally, one pack year represents 20 cigarettes smoked daily for one year. However, pack year numbers can be misleading, because five cigarettes smoked per day for four years is the same as 20 cigarettes smoked per day for one year (one pack year), but the risk of ill health related to smoking is very much greater in the second smoker.
21. Sensitivity analysis is an investigation of potential changes or errors and their impact on the conclusions that have been drawn from a model. It commonly involves a repeat of the primary analysis using alternative decisions or ranges of values where decisions appear arbitrary or unclear.
22. Committee on Health Risks of Exposure to Radon, 1999. Health Effects of Exposure to Radon: BEIR VI. Washington: National Academy Press.
23. Darby et al., 2004. Radon in homes and risk of lung cancer: collaborative analysis of individual data from 13 European case-control studies. British Medical Journal, 330(7485), 223–228.
24. Committee on Health Risks of Exposure to Radon, 1999. Health Effects of Exposure to Radon: BEIR VI. Washington: National Academy Press.
25. Health Canada, 2006. Report of the Radon Working Group on a New Radon Guideline for Canada.
26. Specificity is the ability to correctly identify the proportion of actual negatives that are correctly identified as such. According to the working group, specificity is said to be established when a single accepted cause produces a specific effect.
27. Kataoka et al., 2012. Protective effects of radon inhalation on carrageenan-induced inflammatory paw edema in mice. Inflammation, 35(2), 713–722.
28. Erickson, B.E. (2007). The therapeutic use of radon: a biomedical treatment in Europe; an “alternative” remedy in the United States. Dose-Response : A Publication of International Hormesis Society, 5(1), 48–62.
29. Taubes, G., 1995. Epidemiology faces its limits. Science, 269(5221), 164–169.
30. Alavanja et al., 2000. Letters to the Editor RE: “Residential radon gas exposure and lung cancer: The Iowa radon lung cancer study.” American Journal of Epidemiology, 152(9), 895–896.
Robyn Reist, PEng (engineering physics), worked in radiation protection at Cameco Corporation in Saskatchewan from 2010 to 2017. She is now pursuing an MSc in community and population health science with a focus on ergonomics/human factors, at the University of Saskatchewan. She has enjoyed learning about and applying academic research processes, but she plans to return to industry when she completes her MSc, with a new perspective on why injuries, accidents, and ill health occur in workplaces and how to prevent them.
Robyn Reist, ingénieure physique, a travaillé en radioprotection pour Cameco Corporation en Saskatchewan de 2010 à 2017. Elle étudie présentement à la maîtrise en science de la santé de la population et de la communauté avec spécialisation en ergonomie/facteur humain à l’Université de la Saskatchewan. Elle aime approfondir ses connaissances sur les processus de recherche académique et les mettre en pratique. Toutefois, elle compte retourner en industrie une fois sa maîtrise terminée. Elle apportera ainsi une nouvelle perspective quant aux raisons pour lesquelles des blessures, des accidents et des maladies surviennent au travail, et sur les façons de les prévenir.