Cigarette Smoking Increases Risk for Breast Cancer in High-Risk Breast Cancer Families¹

The data linking cigarette smoking with breast cancer are conflicting. The last published review article on the topic concluded that there was little evidence to suggest that cigarette smoking materially increased risk (1). Most of the earliest studies focused on the hypothesis that smoking may reduce risk, motivated by a report from MacMahon et al. (2) that urinary estrogen levels of smokers during the luteal phase of the menstrual cycle were reduced compared with never-smokers. With the exception of a few early case-control studies that used biased control groups, most of the published studies have described a weak positive association between smoking and breast cancer risk (1). Studies with at least 500 cases reported since the publication of that review have failed to clarify the issue (3, 4, 5, 6).

In apparent contrast to the findings among women in the general population, a recent study (7) among women who were carriers of mutations in BRCA1 or BRCA2 reported that cigarette smoking of greater than four pack-years conferred an approximate 50% decreased risk for breast cancer (odds ratio, 0.46; 95% CI³, 0.27–0.80). Additional research is needed to confirm this observation. The current report examines the association between smoking and breast cancer in high-risk families that were not restricted to BRCA1 and BRCA2 mutation carriers.

Details of the study design have been published (8). Briefly, the cohort of families was originally established through a consecutive series of 544 breast cancer probands ascertained between 1944 and 1952 at the University of Minnesota to examine the influence of childbearing, breastfeeding, and hereditary susceptibility on the risk of breast cancer. The probands resided throughout the state of Minnesota with a minor concentration in the cities of Minneapolis and St. Paul and were unselected for age or family history. The initial study was restricted to Caucasian women because very few minority women were available for meaningful analysis (9). At the baseline period, complete pedigrees of the proband’s parents, siblings, offspring, aunts, uncles, nieces, and nephews were constructed. Reports of cancer among family members were validated with medical records, pathology reports, and death certificates.

Between 1991 and 1996, a historical follow-up study was conducted (8). Of the original 544 families in the cohort at the start of follow-up in 1952, we excluded 40 families because the proband had prevalent breast cancer (diagnosed before 1940) and 42 families because no or very few relatives were alive at baseline. A total of 20 families were lost to follow-up; 10 had no living first- or second-degree relatives, and 6 families refused to participate. Thus, a total of 426 families were studied, representing 92% of those eligible after baseline exclusions.

Pedigrees were updated by genetic counselors according to a structured protocol to reflect births, deaths, marriages, and divorces since 1952 (10). Names, addresses, and phone numbers were obtained on living sisters, daughters, granddaughters, nieces, and women who married corresponding male relatives. For subjects who had died since last contact, identifying information on a suitable surrogate was obtained. The priority for surrogates was, in descending order, spouse, adult offspring, sibling, or parent. Over 95% of the surrogate respondents were spouses or first-degree relatives.

Data on cancer incidence and known risk factors were collected by telephone for female family members over the age of 18 years at the time of follow-up. A sample of 138 breast cancers has been validated against medical records, with only two discrepancies (11). Self-respondents were asked whether they had ever smoked cigarettes (defined as more than 100 cigarettes), the age at which smoking started and stopped, and the average number of cigarettes smoked/day. No information was requested about specific cigarette brands or the use of filters. The only question asked of surrogate responders was whether or not the subject in question had ever smoked cigarettes. The participation rate for the telephone interview was 93%. The percentage of surrogate respondents for cases and noncases was 58% and 30%, respectively.

Of the 426 families studied in the Minnesota Breast Cancer Family Cohort, 166 (39%) had only one family member with breast cancer (the original proband), and 128 (30%) had only two relatives with breast cancer (the proband plus one other relative). To focus the analysis on those families at presumably greatest genetic risk, the current report is based on the subset of 132 families with three or more members affected with breast or ovarian cancer.

The cohort at risk was defined as female family members who were free of cancer at the proband’s date of diagnosis. Follow-up began at age 18 years or when the proband in the family was diagnosed, whichever was later. Follow-up continued until age at breast cancer diagnosis, death, or end of follow up, whichever came first.

RRs and 95% CIs were used as the measures of association between smoking and breast cancer and were estimated using Cox proportional hazards regression (12). Survival was modeled as a function of age, because age is a better predictor of breast cancer risk than is length of follow-up time in this study. Only smoking histories before onset of breast cancer were considered. Analyses were stratified on birth cohort (quartiles) to control for potential cohort effects in smoking and breast cancer incidence. In addition, we accounted for the nonindependence of observations within families by using a robust variance estimate (13).

All of the biological relatives of the initial probands have, by definition, a family history of the disease. However, the degree of relatedness to a woman with breast cancer varies by relative categories. This variation is reflected in the proportion of genes shared ibd. In particular, sisters and daughters share, on average, 50% of their genes ibd with the proband; granddaughters and nieces share 25% of their genes ibd. In addition, the women who married into the families can be expected to reflect the general population at average genetic risk. Therefore, because the possible association between smoking status and breast cancer risk may be expected to differ depending on the relative probability that a subject inherited a high-risk susceptibility gene, we performed analyses within degree of relationship categories. In all of the models, never-smokers were defined as the referent category.

Initially, the association of smoking with development of breast cancer considered only smoking status defined as ever versus never. Because these data were available on both self- and surrogate-responders, they were combined for analysis. Subsequent analyses that explored more detailed aspects of smoking history were restricted to self-respondents. For these analyses, we also examined age at initiation, cigarettes/day, and the number of pack-years smoked in a lifetime (defined as the average number of packs smoked in a day multiplied by the total number of years smoking). The pack-years variable was categorized as never smoked, ≤20 pack-years, and >20 pack-years, based on a cut point at the approximate median of the distribution for ever-smokers. The additional smoking information collected on self-respondents also allowed us to evaluate smoking status and pack-years as time-dependent variables for this group of women. Multivariate models were also fit that allowed for the potential confounding effects of age at menarche, age at first live birth, body mass index, alcohol use, and oral contraceptive use. The addition of these variables did not alter the results. Therefore, only the simpler models are presented here.

The distribution of the 132 families with at least three breast or ovarian cancers by number of cancers, relatives affected, and earliest age at onset is given in Table 1. The family classification reflects data on a total of 2114 relatives. Seventy-six percent of these probands had at least one daughter or sister with breast cancer, and 85% had at least one granddaughter or niece with breast cancer. Roughly half of the families (53%) had a member with onset of breast cancer before the age of 45 years.

Table 2 summarizes the families in terms of relationship to the breast cancer proband, birth cohort, age at interview, and mean age at onset of breast cancer for those subjects who participated directly in the telephone interview. These data include 1369 blood relatives in the 132 high-risk families and 2768 marry-ins from all of the 426 families. As expected, the mean age at interview was higher for relatives closer to the proband (sisters and daughters) than for second-degree relatives (granddaughters and nieces) or marry-ins. In the 132 high-risk families, 113 breast cancers occurred among the blood relatives of the probands, whereas 86 breast cancers developed in the marry-ins from the 426 families. Mean age at onset of breast cancer was similar across relationship categories.

Table 3 presents descriptive data on the smoking history of family members. The prevalence of smoking among direct respondents was slightly lower for sisters and daughters than for nieces, granddaughters, or marry-ins. However, when data from surrogate respondents were included, the differences across relationship categories increased, reflecting cohort trends in tobacco use in this country. Among smokers (self-respondents only), there were no notable differences across relationship categories in age at initiation, cigarettes/day, or pack-years of tobacco exposure.

Analyses were performed to identify risk factors for breast cancer that may confound the association of tobacco use and breast cancer. As shown in Table 4, ever-smokers tended to have a slightly earlier age at menopause and were more likely to have used oral contraceptives and to drink more alcohol-containing beverages than nonsmokers.

The association between smoking status and incidence of breast cancer by relationship to proband is shown in Table 5. Analyses were performed that considered all of the respondents (self- and surrogate-responders combined) and self-respondents only. Among sisters and daughters, those who ever smoked were at 1.8-fold increased risk of breast cancer (95% CI, 1.2–2.7) compared with never-smokers. No association was observed among nieces, granddaughters, or women who married into the families. When the analysis was restricted to women with self-reported data, the association between smoking and breast cancer risk did not change in the nieces and granddaughters or marry-ins. However, the association was strengthened somewhat for the sisters and daughters (RR, 2.4; 95% CI, 1.2–5.1).

Additional aspects of smoking history were also explored. Because these data were not available on surrogates, the analysis was restricted to self-responders. Similar to the previous analyses, risks associated with smoking continued to be limited to first-degree relatives of the probands. No clear pattern with either age at initiation or levels of smoking was observed (data not shown).

To isolate the subset of families with the greatest potential for inherited risk of breast cancer, analyses were then restricted to the highest-risk families. Two definitions of highest risk were used. The first was based solely on the number of relatives affected with breast or ovarian cancer. We selected as a cut point those with five or more cases of breast and/or ovarian cancer; this identified 35 families. Because this approach does not consider family size or age of the relatives at risk, we also defined highest risk based on a standardized morbidity ratio. This expressed the number of observed breast cancers relative to the number expected in the family based on application of age- and race-specific rates from the Iowa Surveillance, Epidemiology, and End Results registry to the family structures. We selected as a cut point those families with two or more excess observed cancers than expected; 38 families were identified. Results of these analyses were similar regardless of the classification used (Table 6). Among sisters and daughters in the 35 families with five or more affected members, those who ever smoked were at 5.8-fold increased risk (95% CI, 1.4–23.9) compared with never-smokers. Smoking status was not associated with significantly elevated breast cancer risk in nieces and granddaughters. Given the reduced number of cases available for this subset analysis, it was not possible to evaluate risk by age at initiation, cigarettes/day, or pack-years of tobacco use. When the standardized morbidity ratio approach was used, the risks to first-degree relatives was attenuated (RR, 2.3; CI, 0.9–6.0).

In this study, cigarette smoking was associated with significantly increased risk of breast cancer among sisters and daughters of probands from families with three or more breast and/or ovarian cancer cases. The increase in risk was 1.8-fold for all of the subjects and 2.4-fold when restricted to self-respondents. Among the subset of 35 families with five or more breast or ovarian cancers, sisters and daughters who smoked were nearly six times more likely to get breast cancer than similar relatives who did not smoke. Smoking-associated risks for more distant relatives were less pronounced. Evidence for a smoking effect was observed among the granddaughters and nieces in the highest risk families, but the point estimate of risk was not statistically significant.

There are at least four possible interpretations of our findings. The first is that the results are real and perhaps a reflection of an interaction between smoking and BRCA1 or BRCA2 mutations. Multiple carcinogens in cigarette smoke are capable of inducing DNA damage. In the absence of functional protein from the BRCA1 and BRCA2 DNA repair genes and perhaps other predisposition genes that also play a role in DNA repair, exposure to these carcinogens may result in accumulation of DNA damage. In time, a tumor suppressor gene such as p53 may be damaged, leading to uncontrolled tumor growth. This model is consistent with the hypothesis that BRCA1 and BRCA2 function as caretaker genes that regulate DNA repair and genomic instability (14). It is likely that a substantial proportion of the participants in the current study carry mutations in BRCA1 or BRCA2, especially the 35 families with five or more cases of breast and/or ovarian cancer. Each of these 35 families has at least a fifty-fifty probability of segregating BRCA1 or BRCA2 mutations (15), suggesting that 18 mutation positive families are enrolled in this study. Another method for estimating the number of mutation positive families is to use population-specific mutation frequency data. The study cohort is predominantly of Scandinavian and Germanic ethnicity reflective of the Minnesota population in the 1950s. Recent studies (16) indicate that 20% of Scandinavian high-risk breast cancer families with three or more cases of breast and/or ovarian cancer carry mutations in these genes, suggesting that as many as 26 of the 132 families may carry mutations. The number of BRCA1 and BRCA2 families can also be predicted using population estimates (17, 18). Given that the original 426 families represent a sequential series of breast cancer cases, we estimate that approximately 15 to 18 families carry BRCA1 or BRCA2 mutations.

The second possible explanation for the result obtained in this study is that smoking is not the relevant exposure but rather some other environmental factor that is closely associated with it. The risks associated with smoking were not explained by any of the other measured risk factors for breast cancer available on the study population. Thus, if this interpretation is true, the relevant risk factor is either dietary in nature or one that has not been identified. This seems unlikely.

A third possible explanation is that the results reflect gene by environment interactions of genes other than BRCA1 or BRCA2. Even if there are 15 to 18 families segregating mutations in these two genes in our cohort, some of the results are based upon patterns of disease in families not segregating BRCA1 or BRCA2 mutations. It is plausible that the genetic predisposition may reflect common genetic polymorphisms in genes that are involved in the biotransformation of procarcinogens in tobacco; e.g., investigators have provided evidence of gene by smoking interactions for N-acetyltransferase 2 (19), N-acetyltransferase 1 (20), and CYP 1A1 (21). The interaction of N-acetyltransferase 2 with smoking on risk of breast cancer has not been replicated in two subsequent studies (22, 23). It would be important to characterize the participants in the current study at these genetic loci.

The fourth possible interpretation of our data is that the observations are simply the result of chance or unknown sources of bias. Although the risk associated with smoking was consistent for self- and surrogate-respondents, the only significant increases in risk were observed among first-degree relatives. One might have expected an association in the same direction, but perhaps of lower magnitude, among second-degree relatives. There was also no evidence for a dose-response relationship with pack-years of tobacco exposure. Other parameters of smoking history, such as age at initiation, were not strongly associated with risk. The magnitude of the relative risk among first-degree relatives is sufficiently high that the potential for such an association to be attributable to chance is reduced.

The data from our study are in apparent contrast with the results from Brunet et al. (7). Their data suggest that individuals with BRCA1 and BRCA2 mutations who smoke cigarettes for >4 pack-years have a reduced breast cancer risk (odds ratio, 0.46; 95% CI, 0.27–0.80) in comparison with those who never smoke. The investigators hypothesized that this finding may reflect an antiestrogenic effect of smoking, thereby reducing the exposure of breast cells to the growth-promoting effects of this hormone. However, the lack of strong empirical data on the association between smoking and endogenous or urinary levels of estrogens (24) makes this a less compelling hypothesis. It is important to note that the current study and the Brunet study (7) cannot be directly compared, because the methods and populations under study are substantially different. In the current study, a genetically undefined high-risk breast cancer population was analyzed, whereas in the study by Brunet et al. (7) only BRCA1 and BRCA2 mutation carriers were included. An important difference is that the current study is able to evaluate all of the cases of breast cancer and is not limited to those cases that survived long enough to be interviewed. Given the evidence that smokers with cancer have a worse prognosis than nonsmokers with cancer have (25, 26), the potential for survival bias in the Brunet study (7) must be considered. Finally, the current report is derived from a population-based sample of families with a high participation rate, and data were collected according to a single protocol. Conversely, the Brunet study (7) is based on a sample of known mutation carriers collected from a number of medical centers. Hence, the denominator of the source population is more difficult to determine, and the participation rates were not cited in their report. In addition, roughly one-third of eligible cases were excluded because of lack of a suitable control. These differences may help to explain the opposing outcomes from the two studies.

This present study is a hybrid of a traditional pedigree study and a cohort study. A distinct advantage of this design is that, unlike most pedigree and many cohort studies, the probands were not selected for early age at onset or family history. All of the breast cancer diagnoses among probands were confirmed by medical record review. In addition, the long-term follow-up (almost 50 years) of the families has been virtually complete, and on average no more than one or two individuals/family were lost to follow-up. Regardless, there are some limitations to our study design that must be considered when interpreting the results. Self-reported breast cancer occurrence and exposure data (including smoking history) were ascertained simultaneously when the historical cohort was updated during the period from 1991 to 1996. However, self-report of breast cancer was found to agree with a medical record and pathologist review in 98% of the cases selected for validation in this cohort. Although there is a potential for differential recall of smoking history by breast cancer status, it seems unlikely that such a bias would be substantial, because we used a standardized telephone questionnaire, and smoking has not been strongly linked to breast cancer. There is also a potential for a survivor bias, which could occur if women with breast cancer who smoked had a poorer survival than the nonsmoking cases did. However, we collected smoking data on over 94% of the deceased family members through surrogate interviews of a close living relative. Our results were generally consistent whether or not surrogate data were used, suggesting that this is not likely to be a strong source of bias in our study.

A dose response to pack-years of smoking was not observed. In fact, among sisters and daughters of probands, lighter smokers were at somewhat greater risk of breast cancer than were heavy smokers. Our results are consistent in this regard with Brunet et al. (7). This apparent paradox may be explained by the hypothesis that in the genetically susceptible, a smaller dose of carcinogen exposure may be sufficient to cause unrepaired DNA damage, leading to initiation or progression of tumor formation. By analogy, similar observations have been reported with regard to genetic susceptibility and lung cancer (27). As the risk of breast cancer was also greater in lighter smoking marry-ins in comparison with heavy smokers, it is also possible that the apparent increased risk and absence of dose response may be attributable to competing mortality in the heavy smoking subset of individuals. Finally, rather than any true biological basis, the lack of dose response could also just reflect chance variation around the true estimate of effect, which may indeed be linear.

In conclusion, the results of this study suggest that women at familial risk for breast cancer are at further and, in some cases, substantially increased risk if they smoke cigarettes. Because smoking is a modifiable risk factor, these women should be advised to avoid smoking to limit their risk for development of breast cancer. This message is critical given published reports (7) that smoking may lower risk of breast cancer among genetic susceptibles. Clearly, there are many reasons to avoid smoking, and perhaps the addition of another reason not to smoke, in the form of this result, will convince current smokers or possible future smokers to avoid the practice.

We thank the study families for their outstanding continued support of this research effort.