Incidence of lung cancer in a large cohort of non-smoking men from Sweden

Abstract

Valid data on the occurrence of lung cancer among non-smokers are sparse and refer mainly to North American populations. The objective of our study is to provide an estimate of the incidence of lung cancer among non-smokers in a large European population. Proceeding from a nation-wide health care program among Swedish construction workers, we estimated the incidence of lung cancer among 143,998 males who reported to be never smokers at 1 or several physical check-ups (average, 2.9 visits) during 1971–92. Cancer incidence data for 1971–95 were obtained through linkage with the virtually complete national cancer registry. Cohort members contributed over 2 million person-years of observation, during which 101 lung cancers occurred. The age-standardized rate was 3.7/100,000 (95% confidence interval [CI] 2.8–4.6). There was a suggestion of a calendar period effect, with rates increasing from 1.5/100,000 in 1976–80 to 5.4/100,000 in 1991–95. A birth cohort effect was also suggested. Rates estimated in this population are comparable to those from North American studies. The increasing rate during the study period might be due to non-smokers who took up the habit, but it might also reflect temporal changes in the effect of other risk factors of lung cancer in non-smokers. © 2001 Wiley-Liss, Inc.

Tobacco smoking is the overwhelming cause of lung cancer in most populations. For this reason, occurrence of lung cancer largely reflects tobacco consumption in previous decades. The importance of tobacco smoking as a cause of lung cancer complicates the study of other determinants of the disease. In particular, studies of non-smokers require a valid and precise definition of non-smoking status, because even a small misclassification of smokers as non-smokers would substantially affect lung cancer rates among the latter. In addition, studies of non-smokers require large populations, given the relative rarity of this disease in the absence of smoking.

For this reason, valid data on the occurrence of lung cancer among non-smokers, in particular among non-smoking men, are sparse,1 and most published results refer to relatively small populations. The available estimates are mainly from American studies and no large studies have been reported from Europe.

We report here an estimate of the incidence of lung cancer in a large cohort of non-smoking men from Sweden who were enrolled in a nation-wide health surveillance scheme of construction workers.

METHODS

The Swedish Construction Industry's Organization for Working Environment Safety and Health (Bygghälsan) provided outpatient medical services to construction workers all over the country during 1969–92.2 The organization was a joint venture between trade unions and the employers' association. It had stationary and mobile units, thus reaching the majority of organized construction workers in Sweden. A main activity consisted of preventive health examinations that were offered to all blue- and white-collar workers in the construction industry through regular personal invitations (every 1–5 year depending on exposure and time period). Beginning in 1971, data from these health check-ups were compiled in a computerized register.

Before each visit, workers filled out a questionnaire, including a detailed smoking history. At the clinic, the answers were checked by a nurse to reduce misunderstanding and inconsistent answers. The quality of data on smoking has been reviewed.3 Perfect concordance between reports of smoking status (at the first and second health examination 2–3 years apart) was found in 89% of 18,593 subjects. Inconsistencies regarding never-smoking status (i.e., subjects who indicated they were current or former smokers in the first questionnaire and never smokers in the second questionnaire) were found in 505 subjects (2.7%). However, they represent 9.0% of 5,632 subjects who reported never smoking at the second questionnaire. After the closure of Bygghälsan, the data were transferred to Umea University, which ensures the maintenance of the database and the continuation of the follow-up. The proposal for our analysis was reviewed and approved by the Bygghälsan Steering Committee. For the purpose of this analysis, we selected a cohort of workers who reported themselves as never smokers at all visits. Never smoking was defined as not having smoked 1 cigarette a day during 1 year. Occasional smokers, however, might be included among never smokers. The cohort comprised 143,998 male workers, out of a total of approximately 390,000 workers included in the register.

The prevalence of never smokers in 1974 was 33.8%, that of ex-smokers was 20.4% and that of current smokers was 38.3% (unknown information for 7.5%). These proportions are comparable to those found in samples of the Swedish male population. The educational level is varying, the majority of workers having had between 7 and 9 years of education.

Each cohort member was identified by the national registration number, a unique personal identifier assigned to all residents in Sweden. Efforts were made to ensure that the registration numbers were complete and valid: the proportion of cohort members with incorrect numbers was less than 0.2%. The registration number was used for linkage to the Cancer Register, the Mortality Register and the Migration Register. The Cancer Register, founded in 1958 and more than 98% complete,4 has coded malignant neoplasms during the entire study period according to the 7th Revision of the International Classification of Diseases (ICD-7). This study is restricted to the incidence of lung cancer (ICD-7: 162.1). Each cohort member contributed person-years of observation from the date of the first registered visit until the date of diagnosis of lung cancer, death, migration, or the closing date of 31 December 1995, whichever came first.

We analyzed the incidence of lung cancer in 15 categories of age (15–19, 20–24,…, 80–84, 85+), 5 categories of calendar period (1971–75,…, 1991–95) and 6 categories of birth cohort (before 1915, 1915–19,…, after 1934). We calculated age-standardized incidence rates and their standard errors based on the world standard population.5 The presence of period effects and cohort effects were assessed separately by fitting age-period and age-cohort Poisson regression models and calculating the difference in deviance with respect to the model including only terms for age.6

We stratified cohort members according to the number of visits at which the workers reported their non-smoking status, to address the possibility of misclassification due to changes in smoking habits after the last registered visit. In addition, we stratified cohort members according to possible exposure to asbestos. The latter exposure was defined according to a job-exposure matrix developed for a previous analysis of lung cancer risk according to exposure to asbestos and man-made mineral fibers.7 The matrix covered the jobs held by construction workers until 1984 and permitted to classify 58% of cohort members.

RESULTS

Cohort members contributed a total of 2,059,961 person-years of observation. The mean age at entry was 30.4 years (range 14–82) and the mean duration of follow-up was 14.3 years (range 0.01–24.5). The average number of visits registered for each cohort member was 2.9 (range 1–12). During the follow-up, 101 lung cancers occurred, yielding a crude incidence rate of 4.9/100,000 (95% CI 4.4–6.4). The age-standardized rate was 3.7/100,000 (95% CI 2.8–4.6). There were no cases before age 35. Age-specific rates increased steadily after that age, up to 64.5/100,000 above age 84 (Table I). Only 1 case occurred before 1976; age-standardized rates in subsequent quinquennia increased from 1.5/100,000 in 1976–80 (95% CI 0.1–2.9) to 5.4/100,000 in 1991–95 (95% CI 4.5–6.3), suggesting a period effect. Corresponding rates truncated to the age group 35–64 were 17.4/100,000, 24.1/100,000, 21.9/100,000 and 48.3/100,000. The results of the analysis by cohort of birth also suggested a cohort effect. For example, rates at age 65–74 were 13.5/100,000, 31.7/100,000 and 56.8/100,000 in those born before 1915, in 1915–24 and in 1925–34.

Table I. Age-Specific Morbidity Rates of Lung Cancer Among Non-Smoking Swedish Construction Workers

Age group	Person-years	N cases	Rate1	95% CI2
1 Per 100,000 person-years. 2 CI, confidence interval.
35–39	232,612	2	0.9	>0–2.1
40–44	207,231	4	1.9	0.01–3.8
45–49	182,277	4	2.2	0.04–4.3
50–54	151,430	5	3.3	0.4–6.2
55–59	131,634	9	6.8	2.4–11.3
60–64	115,774	15	13.0	6.4–19.5
65–69	90,286	26	28.8	17.7–39.9
70–74	58,616	18	30.7	16.5–44.9
75–79	29,827	12	40.2	17.5–63.0
80–84	10,267	5	48.7	6.0–91.4
85+	1,550	1	64.5	0–191.0

To investigate period and cohort effects in more detail, these were plotted with respect to age (Figs. 1, 2). These graphs indicate the presence of both period and cohort effects in the increasing lung cancer mortality, although it is not possible to identify which of the two is predominant.

Figure 1. Age- and period-specific rates of lung cancer.

Figure 2. Age- and birth cohort-specific rates of lung cancer.

The stratification according to number of visits resulted in a higher rate for workers with only 1 visit (ASR during the whole period 4.7/100,000) as compared to workers with 2–3 visits (4.0/100,000) or 4 or more visits (2.1/100,000) and this despite the fact that person-years after the 3rd and 4th visit mainly accumulated in the later part of the follow-up. After stratification, the difference according to number of visits was greater during the last quinquennium of follow-up than in previous periods. Stratification of cohort members according to possible asbestos exposure resulted in similar rates in the 2 groups: cohort members ever employed in jobs classified as possibly exposed (n = 37,369) experienced an ASR of 3.4/100,000, whereas the ASR among cohort members never employed in such jobs (n = 46,590) was 3.7/100,000.

DISCUSSION

The age-standardized rate of lung cancer (based on the world population) in this cohort of non-smoking Swedish construction workers was 3.7/100,000. The crude rate was 4.9/100,000. A comparison of our rates with those (mainly based on mortality) reported in other studies of lung cancer in non-smokers8–14 is complicated by the fact that in several studies only crude rates were reported (or can be computed) and in other studies the populations used to derive standardized rates differed. For 3 American studies, based on mortality, however, age-specific rates were reported and they are plotted in Figure 3 together with our results. It is interesting to see that rates seem fairly homogeneous: the truncated rates (age 35–84, use of world standard population) were 3.1/100,000 in the cohort of U.S. Veterans,8 5.0/100,000 in the American Cancer Society CPS-II study14 and 3.4/100,000 in our study. Our results are also in agreement with data reported during 1986–92 from the Missouri Cancer Registry, in which yearly age-adjusted incidence rates varied between 4 and 7/100,000.1 Relatively few data are available from Europe and they only concern rates not adjusted for age.9, 12, 13

Figure 3. Comparison of age-specific rates of lung cancer among non-smokers. *, 75+ in CPS-I study. Squares, US veterans (Kahn 8); diamonds, CPS-I (Garfinkel 10); circles, CPS-II (Burns et al. 14);triangles, this study.

Our results suggest a similar lung cancer rate among non-smokers in Sweden as in the United States: the lower nation-wide incidence of lung cancer in Sweden, as compared to other Western countries,15 is attributable to a lower tobacco consumption16 rather than to other environmental or genetic factors.

The fact that we observed a higher rate among cohort members with only 1 medical visit speaks in favor of a possible bias from assessment of non-smoking status at only one point in time, as it was done in most of the previous studies. On the other hand, workers who participated more zealously in these preventive health check-ups may have been generally more health conscious than sporadic visitors.

It is of interest to note that in our population, 40% of cases were adenocarcinomas, 19% were squamous cell carcinomas, 29% were small cell carcinomas and 12% had other histologies (information was missing for 1 case), which is similar to the distribution in another series of male non-smoking cases from Europe.17

The increase in rate with increasing age (Table I) is broadly consistent with a power function of the form rate ∼ age ** C. Our data fit better a cubic than a quadratic form, whereas models with higher powers (C = 4 or 5) do not converge because of small number in the extreme age groups.

We found evidence of an increase in the incidence of lung cancer among non-smokers during the study period. This result might be explained by an increasing number of cohort members who took up smoking after the last visit. On the other hand, we cannot exclude a variation in risk factors of lung cancer other than smoking. Despite the limitations of analyses on period and cohort effects, our results suggest a stronger effect of calendar period than birth cohort, which would point toward exposures which affect the population equally, such as occupational carcinogens and air pollution, rather than exposures which affect differently people born at different times, such as spousal passive smoke. There is no evidence of an increased prevalence of smoking among women during the study period. It should be stressed, however, that an increasing taking up of smoking during the study period would also result in a calendar period effect. The prevalence of smoking among men declined during the study period, suggesting that uptake of smoking in adulthood is an unlikely explanation of our findings.

Our study has a number of limitations. The cohort is not a random sample of the Swedish population. Participation of construction workers in the health program was almost universal, however and no selection is likely to have occurred on factors possibly related to subsequent lung cancer risk. Although construction workers may constitute a selected group with a low prevalence of chronic pulmonary conditions such as e.g., asthma, conceivably resulting in a moderate healthy worker effect, we have little reason to believe that these workers were more exposed to protective dietary factors (fruit and vegetables) compared to samples examined in previous studies.

Exposure to occupational carcinogens might have increased the rate of lung cancer in our cohort. However, we addressed the potential confounding of the most likely candidate, asbestos and we found no evidence in favor of it.

In conclusion, our study provides the so far most precise estimate of the incidence of lung cancer in a non-smoking male population from Europe. We found evidence of an increased incidence between 1971 and 1995, which might be best explained by a calendar period effect.

Acknowledgements

We acknowledge the work of those who designed and organized the Bygghälsan program, in particular Drs. A. Englund and G. Engholm.