Breast Cancer Risk Among Women Who Start Smoking as Teenagers

The relationship between smoking and the risk of breast cancer has been unclear for several decades, and whether or not young women should be warned about smoking as a possible cause of breast cancer remains controversial (1-4). In 1982, MacMahon et al. (5) first suggested a protective effect, whereas Hiatt and Fireman, 4 years later, proposed a detrimental effect of smoking on breast cancer risk (6). Since then, studies have continued to show positive, inverse, or null associations (4, 7-9).

In 2002, a pooled analysis of 53 epidemiologic studies found that the positive relationship between smoking and breast cancer was likely due to confounding by alcohol consumption, an established cause of breast cancer (10). However, this study used crude smoking measures, e.g. ever versus never smokers and also including passive smokers in the reference group, possibly diluting a weak effect. From a public health perspective, even a small increase in the risk of breast cancer among women who smoke may account for a substantial number of cancer cases in populations where the prevalence of smoking is high.

In 1982, Russo et al. (11) hypothesized that the mammary tissue is particularly susceptible to carcinogenic exposures during the years from pre-puberty to the first birth, when the mammary cells differentiate. Therefore, we examined the association between cigarette smoking and breast cancer risk in a large population-based cohort of women who were teenagers when smoking prevalence increased markedly among young Scandinavian women.

The Norwegian-Swedish Women's Lifestyle and Health Cohort Study was initiated in 1991/1992. In Norway, a nationwide random sample of 100,000 women, born between 1943 and 1957, was drawn from the National Population Register at Statistics Norway. In Sweden, a random sample of 96,000 women, born between 1943 and 1962, residing in the Uppsala Health Care Region, was drawn from the National Population Register at Statistics Sweden.

All women received a letter inviting them to participate in the study. The letter requested that they provide written informed consent, and contained a comprehensive questionnaire that was to be completed and returned in a prestamped envelope. The common set of questions included detailed assessment of smoking habits, alcohol consumption, contraceptive use, reproductive history, history of breast cancer in the mother and sister(s), height and current weight [allowing us to calculate body mass index (BMI) as weight in kilograms divided by the square of height in meters], and other aspects of lifestyle habits. In both countries, the national Data Inspection Boards and the regional Medical Ethical Committees approved the study. All women gave written informed consent to participate.

The questionnaire elicited information on current and previous smoking history, and asked if the women lived with a smoker or had done so during childhood. We categorized women who had never smoked, but been exposed to passive smoke at home as “passive smokers” and those reporting neither as “never smokers”.

We categorized the ever-active smokers according to current and past smoking status, smoking duration; number of cigarettes smoked daily, pack-years of smoking (i.e., number of cigarettes smoked per day, divided by 20, multiplied by the number of years smoked), and years of smoking latency (i.e., age at cohort enrollment minus age of smoking initiation). We further separated them into three descriptors of timing of smoking initiation: (a) age they started smoking, (b) whether they started smoking before or around menarche, and (c) before their first birth.

We classified women who had smoked for at least 20 years as long-term smokers, and calculated their average smoking duration, according to the three descriptors of timing of smoking initiation. We categorized past smokers according to smoking recency, i.e., years since they stopped smoking.

Women who reported a natural menopause or a bilateral oophorectomy at cohort enrollment were considered postmenopausal, regardless of age at hysterectomy, or use of hormone replacement therapy. Women were classified as postmenopausal when they reached 50 years of age during follow-up. We calculated average daily consumption of alcohol in grams based on the content of pure alcohol in the different sorts of beverages among drinkers. Women who reported to be teetotalers, and those answering “seldom” or “never” in the frequency table, are from hereon referred to as nondrinkers. Their alcohol consumption was set to 0.

In Norway, 57,584 (57.6%) and in Sweden 49,259 (51.3%) women returned completed questionnaires, resulting in an overall response rate of 54.5%. The cohort data were linked to the national registries of cancer and statistics in Norway and Sweden, to identify all incident cancer cases and deaths/emigrations, respectively. Woman-years were calculated from the start of follow-up to the date of diagnosis of primary invasive breast cancer, the date of emigration, death, or the end of follow-up, i.e., December 31, 2000, whichever occurred first.

Among the 106,841 women included, 789 emigrated and 1,360 died during follow-up. We excluded 5 women due to missing vital status information (whether they were alive, dead, or emigrated) in the available register files, 15 women who had emigrated or died before the start of follow-up, and 1,681 women who were diagnosed with any invasive cancer prior to the start of follow-up for a total of 1,701. We excluded 3,042 women with no information on smoking history, leaving 102,098 subjects in the analytic cohort.

We obtained crude breast cancer incidence rates by dividing the number of cases by the total number of woman-years in that exposure category. The rates were then age-adjusted to the world standard population (12). We estimated the relative risk (RR) of breast cancer associated with different measures of smoking exposure as described by Cox (13). We tested for trends across categories of smoking variables by assigning equally spaced scores to the categories and treating the variable as continuous in the analyses. The trend test analyses did not include passive smokers. Unless otherwise specified, the reference group is designated as never smokers throughout the paper.

Potential confounding variables for the smoking and breast cancer association were examined and, those of importance, i.e., age at enrollment, age at menarche, age at first birth, number of children, menopausal status, family history of breast cancer, hormonal contraceptive use, consumption of alcohol, and BMI, were included in the final models. Age at first birth (<21, 21-24, 25+) and parity (0, 1, 2, 3+) were considered as a set of indicator variables. The multivariate analyses presented here are based on 91,125 women with information on the above listed variables. We repeated all analyses excluding 60 women diagnosed with breast cancer during the first 12 months of follow-up. The RR estimates obtained in this way were not materially different from the results presented in the paper (data not shown).

We conducted analyses on the entire study population, among long-term ever smokers, among nondrinkers, and separately for each country. We also compared the RR of breast cancer among long-term smokers stratified by country and by two categories (high/low risk) of established breast cancer risk factors. Subsequently, we conducted analyses including passive smokers in the reference group. Because smoking influences BMI and age at menopause, we also conducted analyses excluding BMI and age at menopause from the models, one by one, and both at the same time, as well as setting menopause at 54 years for never smokers and 52 years for current smokers.

We did the Cox proportional hazards analyses with the PHREG procedure in the SAS statistical package (14). We entered multiplicative terms between smoking and possible effect modifiers in the proportional hazards model to evaluate interaction, and tested for heterogeneity between strata with Wald χ² statistics. RRs are given with 95% CIs. Results were considered as statistically significant if the P value was 0.05 or less, and 95% CIs are reported throughout the paper. All P values are two-sided.

During the 924,768 person-years of observation, 1,240 incident cases of primary invasive breast cancer, confirmed by histology, were identified. Altogether, 64% of the women were reported to be ever smokers, whereas 13% were classified as never smokers. Table 1 shows that Norwegian women were more likely to be past or current smokers than Swedish women (P < 0.001).

Table 2 shows that the distribution of selected characteristics varies according to smoking status. Current smokers differ from past smokers on all the listed variables in the table (all P values <0.001), except for family history of breast cancer (P = 0.77). The table shows that current and passive smokers also differ significantly from never smokers on most of the selected characteristics.

At enrollment, more than half of the past smokers had quit smoking five or more years ago. All the measures of smoking exposure, i.e., time of smoking initiation, smoking intensity, and duration, differ between current and past smokers in the cohort (all P values <0.001). Altogether, 51% among current and 63% among past smokers were reported to be light smokers, i.e., smoking <10 cigarettes per day. Among nondrinkers, 49% among current and 60% among past smokers were reported to be light smokers. Among current smokers, 69% of the nondrinking and 70% of all women were reported to be long-term smokers (P = 0.34).

The age-adjusted incidence rates of breast cancer were 114 per 100,000 woman-years among never smokers, 120 among past-smokers, and 135 among current smokers. The corresponding rate among women who had been exposed to passive smoking at home was 134 per 100,000 woman-years. A multivariate model based on 1,130 cases with complete information on the potential confounders yielded a nonsignificant increased RR for past 1.15 (95% CI, 0.94-1.41), current 1.17 (95% CI, 0.95-1.45), and passive 1.21 (95% CI, 0.98-1.50) smokers compared with never smokers. An ordinal trend test across three smoking groups (never, past, and current) was not statistically significant. We also collapsed past and current smokers and used both passive and never smokers as the reference group. This model yielded a RR estimate for ever smokers of 1.0 (95% CI, 0.98-1.50). We further examined the association between the time since subjects quit smoking (and different combinations of childhood and current exposures at home among passive smokers) and breast cancer risk. No meaningful results were revealed (data not shown).

Table 3 shows the RR of breast cancer according to various measures of smoking exposure among current and long-term smokers compared with never smokers. In the analysis including all women, current smokers who began smoking before 15 years of age, those who smoked >10 cigarettes a day, or had smoked at least 20 pack-years all had increased risk. Trend tests across the four exposure categories (including the reference group) for number of cigarettes smoked daily (P = 0.03), duration of smoking (P = 0.05), and number of pack-years (P = 0.01) were significant. We observed even stronger associations between smoking and breast cancer risk when the models included only long-term smokers. Those who began smoking before or around menarche had a 39%, and those starting before their first birth had a 27% statistically significant increased risk of developing breast cancer. When these models were restricted to nondrinkers, women who started smoking before their first birth, or had a smoking latency period of >25 years, presented a significantly increased risk of developing breast cancer by about 75%. The results among the nondrinkers are based on a small number of cases, and have wide confidence intervals (Table 3).

We repeated the analyses in Table 3 including passive smokers in the reference group. The associations stayed in the same directions, but became weaker, and only the trend tests for number of pack-years and latency remained statistically significant among current and long-term smokers (all P values <0.05; data not shown). The risk estimates among current smokers shown in Table 3, with the corresponding trend tests, stayed basically the same when we did country-specific analyses, and also when we excluded both the menopausal status and BMI variables from the analyses. Excluding only menopausal status, resulted in weaker dose-response associations, whereas exclusion of BMI yielded a stronger association between smoking and breast cancer risk (data not shown). In the multivariate analyses, we also obtained stronger associations when we set a different age at menopause during follow-up for never (54 years) and current (52 years) smokers instead of the same age (50 years) for all women. For this model, the trend tests for smoking initiation [age at start (P = 0.04), before or around menarche (P = 0.08), or before first birth (P = 0.07)] achieved significant or borderline significant results.

Among long-term smokers, those who started before the age of 15 had a 20% nonsignificant increased risk when those who started between ages 15 and 19 were the reference group. The former group had on average smoked 1.4 years longer than the comparison group (P = 0.001). Likewise, those who started before or around menarche had smoked 1.2 years longer (P = 0.001) and had a 15% increased risk compared with long-term smokers who started after menarche. Similarly, long-term smokers who started before their first birth, had smoked 0.8 years longer (P = 0.001) and had an almost 40% increased risk when the reference group was those who started smoking after their first birth.

The RR of breast cancer among ever smokers compared with never smokers were the same in Norway and Sweden when stratified by country, i.e., 1.24 (95% CI, 0.90-1.70). Table 4 shows the RR estimates among long-term smokers compared with never smokers, after stratification by different risk factors for breast cancer. All risk estimates, except for menarche before age 13, were above unity. The test for heterogeneity was not significant for any of the risk factors. The results from Table 4 stayed materially the same when we included passive smokers in the reference group or did country-specific analyses (data not shown).

This study finds that women who start smoking prior to menarche or a first birth and continue to smoke for at least 20 years have a small increase in breast cancer risk. A causal interpretation of these results is supported by the presence of a consistent dose-response association between various measures of smoking exposure and breast cancer risk. Furthermore, nondrinkers experience a similar increase in risk as drinkers. Our study shows no evidence that family history of breast cancer, menopausal status, or any other established breast cancer risk factor modifies the association with smoking. In contrast, women who have been smoking for 20 years or more, but started after their first birth, do not experience an increased breast cancer risk.

Our study has several strengths. The smoking histories were obtained at enrollment and, hence, are not subject to recall bias. We have the ability to separate current and past smokers, and to exclude women exposed to passive smoking at home from the reference group. We have a large proportion of current smokers at enrollment, and have the ability to stratify long-term smokers according to descriptors of smoking initiation, and compare the different groups of long-term smokers. We have detailed information on, and can control for established breast cancer risk factors: many of which vary according to smoking status. Additional strengths include the large size of the cohort, that it is sampled from a general population, and has virtually complete follow-up. The cohort shows the expected association between oral contraceptive use and breast cancer, and between hormonal contraceptive use and ovarian cancer (15, 16). We have shown that the study participants are similar to nonparticipants with respect to several breast cancer risk factors (17). Furthermore, the cumulated incidence rates during follow-up for all cancer sites as well as for breast cancer sites are almost identical as those reported to the national cancer registry in the same period, suggesting that our findings may be valid in the general population (18). Indeed, the smoking habits among our Norwegian and Swedish women reflect known smoking patterns in the respective countries.

Our study has several limitations. We have only crude and limited information about passive smoking. Also, because time of smoking initiation and duration are highly correlated, in this relatively young cohort, it is very difficult to separate the effects. Furthermore, we do not have updated information on changes in smoking habits, hormonal contraceptive use, hormonal replacement therapy, BMI, menopausal status, and alcohol consumption during follow-up, nor lifetime alcohol consumption. However, few Scandinavian women start to smoke after the age of 30, which is the lowest age of enrollment in our study. If current smokers quit smoking during follow-up, this may have diluted our results. We already show that by setting the same age at menopause for smokers and never smokers, we have biased the displayed results towards the null. In a recent study by Horn-Ross et al. (19), elevated breast cancer risk was most evident for recent drinking with no clear pattern for consumption during earlier periods of life. The lack of lifetime alcohol consumption in our study may still be a concern. Because the distribution of established risk factors for breast cancer vary according to smoking status in our cohort, unknown risk factors may do the same. There may be some residual confounding in both directions due to the factors listed above.

Due to smoking trends in Norway and Sweden, women in our cohort have a higher smoking prevalence during their teenage years than women included in the three previous population-based studies from the Scandinavian countries (20-22). Nevertheless, in our Norwegian-Swedish case-control study including women <45 years old at the time of the data collection in 1985, we did find a 30% nonsignificant increased breast cancer risk among those who started smoking before 15 years old (20). Similarly, the Swedish follow-up study, where the youngest women were 18 years old at the time of the data collection in 1963, revealed a 20% nonsignificant increased risk among those who began smoking before the age of 19 (22). The Norwegian follow-up study, where the youngest women were 35 years old at enrollment in the mid-1970s, found no association with breast cancer risk (21).

To date, 13 cohort studies (6, 21, 23-32) have examined the association between smoking and breast cancer risk. The results of these studies are mixed, but generally show no association. However, the four most recent studies and the only ones including >500 breast cancer cases do find a positive association with either long-term smoking (30), smoking prior to a first birth (29), or both (31, 32).

In contrast to the women in the studies by Egan et al. (29) and Terry et al. (30), our women were aged <50 years at enrollment, thus belonging to birth cohorts with different smoking habits than the two North American studies. The most recently published studies (31, 32) recruited participants during the 1990s, as we did. However, there were only 5% and 13% current smokers at enrollment in the California Teachers Study (31) and in the Nurses' Health Study II (32), respectively. In comparison, we have 28% current smokers in our sample from the general population. This may explain the consistency of our results.

The present study shows a consistent dose-response, which enhances the biological plausibility of an increased breast cancer risk due to smoking. Plasma lipoproteins may transport the carcinogens in the tobacco smoke to the breast (33) where they can be stored in breast adipose tissue (34, 35) and could then be metabolized and activated by human mammary epithelial cells. Furthermore, the breast tissue of smokers have a higher prevalence of smoking-specific DNA adducts and p53 gene mutations compared with that in nonsmokers (7). This all supports the likelihood of a positive association between cigarette smoking and breast cancer risk. The increased risk associated with smoking prior to a first birth is in accordance with data from epidemiologic and animal studies (7, 11, 36, 37). One important result from our study is that we find no increased risk among women who initiate smoking after the first birth. This supports the notion that breast tissue becomes more resistant to the carcinogens in tobacco smoke after differentiation.

Our study finds that women who start smoking at young ages, and continue to smoke for at least 20 years increase their breast cancer risk. The fact that teenagers who smoke today tend to initiate smoking at earlier ages than in previous generations may have implications for targeted efforts to reduce breast cancer risk. Teenage girls and adult women should be given yet another reason to avoid, and to quit smoking.