Alcohol, Physical Activity, Smoking, and Breast Cancer Subtypes in a Large, Nested Case–Control Study from the Norwegian Breast Cancer Screening Program

This article is featured in Highlights of This Issue, p. 1677

Breast cancer is the most common cancer and the leading cause of cancer-related death among women worldwide (1). It has been suggested that modifiable risk factors account for more than one-third of postmenopausal breast cancers in the United States (2).

We have previously reported that hormonal variables, such as parity and hormonal therapy, were associated only with luminal A–like breast cancer (3). Therefore, modifying these hormonal variables would not be expected to lead to reductions in risk of triple-negative or HER2-positive breast cancers. An important question is therefore whether there are other modifiable risk factors that are associated with triple-negative or HER2-positive cancers.

Alcohol consumption, physical activity, and smoking are modifiable risk factors that have been associated with breast cancer overall to some or larger degree (4–6). Several studies have investigated the association between alcohol (7–18), smoking (14, 19–25), physical activity (26–36), and subtypes of breast cancer. A large meta-analysis suggested stronger positive associations between alcohol and estrogen receptor (ER)-positive tumors and weaker positive associations between alcohol and ER-negative tumors (18). There is some evidence that smoking increases risk of hormone receptor–positive [ER-positive and progesterone receptor (PR)-positive tumors] or luminal subtypes (14, 20, 23, 25). However, studies may have been too small to identify associations with triple-negative (ER⁻PR⁻HER2⁻⁾ or HER2-positive cancers. Greater risk reduction has been reported for ER⁻PR⁻ than for ER⁺PR⁺ breast cancer and physical activity and physical activity later compared with early in life (30). However, findings are not entirely consistent (32–34, 37), and it remains unclear whether physical activity protects against all breast cancer subtypes as defined by the expression of ER, PR, and HER2.

A challenge in previous studies has been that the subtypes have not been consistently defined. Often all ER⁻, HER2⁻ tumors have been defined as “luminal A” tumors regardless of whether PR was positive or negative and luminal A–like subtype have included cases considered luminal B–like HER2 negative. However, clinical criteria have suggested that the ER⁺, PR⁻, HER2⁻ tumors may represent a particular type of luminal B–like tumors (38). We wanted to use the definition suggested at the St. Gallen meeting in 2013 (38), and divide breast cancer subtypes into five different subtypes, luminal A–like (ER⁺PR⁻HER2⁻), luminal B–like HER2-negative (ER⁺PR⁻HER2⁻), luminal B–like HER2-positive (ER⁺PR⁺/PR⁻HER2⁺), HER2-positive (ER⁻PR⁻HER2⁺), and triple-negative (ER⁻PR⁻HER2⁻) cancer. Although this was a clinical subdivision, defined to better identify a subgroup of women that could benefit from chemotherapy, we hypothesized that perhaps this would also help better identify a group of luminal A–like cancers with similar etiology.

This is the first study to investigate the association between alcohol intake, smoking, and physical activity and subtypes of breast cancer in a large Norwegian study. We therefore took advantage of the nationwide Norwegian Breast Cancer Screening Program, and conducted a nested case–control study within this cohort.

The methods are described in detail previously (3). In brief, the Cancer Registry of Norway (CRN) is a population-based registry that has collected data on cancer in Norway since 1952. The registry is 98.8% complete (39). CRN is also responsible for the administration of the Norwegian Breast Cancer Screening Program (40). As a part of this screening program, invitations are sent to all women ages 50–69 years old to undergo a two-view mammography screening (standard cranio-caudal and mediolateral views) every two years. The average attendance rate in each round is about 75%. Women who attended the mammographic screening during 2006–2014 were asked to complete a questionnaire on a number of standard breast cancer risk factors before age 50 and another questionnaire on current exposure variables at subsequent screenings.

Because of the short follow-up (38% of the cases was detected in the first screening round), and because we were concerned, we needed to define equivalent time windows of exposure in controls as cases, we chose a case–control design. We conducted a case–control study nested within the study cohort. Information on cancer cases was obtained through linkage to the CRN records using a unique 11-digit personal identification number. Eligible women had no history of invasive cancer (except nonmelanoma skin cancer) or ductal carcinoma in situ of the breast prior to January 1, 2006. Participants who fulfilled these criteria and had completed the questionnaires were eligible for inclusion in the current study cohort (N = 344,348).

Cases were defined as women diagnosed with a first occurrence of invasive breast cancer (ICD10: C50) during 2006–2014, with information on ER, PR, and HER2 receptor status (see below), regardless of screening history. Each breast cancer case was individually matched to five controls on year of birth (±3 years) and year of last screening before breast cancer diagnosis (±3 years). Controls had to be cancer free, alive, and resident in the country at the time of the case's diagnosis.

The study was approved by the Regional Committee for Medical and Health Research Ethics in the South-East Health Region of Norway.

Information on ER, PR, and HER2 status was assessed by IHC and extracted from pathology reports submitted to the CRN. Tumors were classified as being ER⁺ if ≥ 10% reactivity from 2006 to January 2012, and if ≥1% reactivity from February 2012 onwards. The latter threshold was introduced as a result of change in treatment protocols of patients in the clinics in Norway. We used these official thresholds. PR⁺ tumors were defined as those with a reactivity of ≥ 10% throughout the study period. HER2 expression status was determined at each laboratory with IHC. Cases with no (0) or weak (1+) immunostaining were classified as HER2⁻, whereas cases with strong immunostaining (3+) were defined as HER2⁺. In situ hybridization was used to confirm the HER2 status if IHC yielded moderate staining (2+) results. If IHC was 2+ and FISH, chromogenic (CISH), or silver in situ (SISH) were missing, or if IHC was missing but FISH, CISH, or SISH were positive, the tumor was classified as HER2⁺. If IHC was 2+ and FISH, CISH, and SISH were negative, the tumor was regarded as HER2⁻.

Breast cancer cases were categorized according to subtype using a modified version of the classification of clinically defined subtypes proposed at the St. Gallen meeting in 2013 (38). Of the 4,952 breast cancer cases with no missing information on confounders, 550 cases had unknown hormone receptor (i.e., ER and/or PR) and HER2 status or could not be classified into subtypes. Of the remaining 4,402 breast cancer cases, 2,761 were classified as luminal A-like (ER⁺PR⁺HER2⁻), 709 as luminal B-like HER2 negative (ER⁺PR⁻HER2⁻), 367 as luminal B-like HER2 positive (ER⁺PR⁻/PR⁺HER2⁺), 204 as HER2 positive (ER⁻PR⁻HER2⁺), and 361 as triple negative (ER⁻PR⁻HER2⁻). As we did not have Ki-67 results, we conducted a sensitivity analysis whereby grade was added to the luminal subtype definitions in an attempt to better separate out these subtypes (38). In this analysis, luminal A–like subtype was defined as ER⁺PR⁺HER2⁻, low or medium grade, luminal B–like HER2-negative as ER⁺PR⁻HER2⁻, high grade, and luminal B–like HER2-positive as ER⁺PR⁺/PR⁻HER2⁺, any grade. This sensitivity analysis showed that the results remained largely the same and thus only the results without information on grade are presented here. Furthermore, we ran a sensitivity analysis including luminal B–like HER2-negative in the luminal A–like subtype, and another sensitivity analysis to determine whether the cases with no subtype information differed from those with luminal A–like subtype.

The observed subtype distribution in our study—luminal A–like (63%), luminal B–like HER2 negative (16%), luminal B–like HER2 positive (9%), HER2 positive (5%), and triple negative (8%)—was similar to that found in white U.S. women (41).

Data on the exposures of interest were extracted from the questionnaires completed at the last screening prior to breast cancer diagnosis for the cases, and the corresponding round for controls. If a woman's questionnaire, or data on relevant variables, were missing, information from the questionnaire completed in the previous screening round was used. Approximately 16.5% of the questionnaires were replaced by a questionnaire from the previous round. The primary exposures of interest were alcohol consumption, physical activity, and smoking habits. Participants were asked to recall these exposures at different time periods (before 50 years and currently). Women were asked about the amount of beer, wine, or liquor consumed in glasses per week. The amount of total alcohol intake was estimated assuming 14 grams of ethanol per glass of liquor, 20 grams per 0.5 L of beer, and 12 grams per glass of wine. To make the number easier to relate to, we converted the alcohol consumed per week into glasses per week, assuming every glass would have the same alcohol content as a glass of wine (12 g). The tables therefore contain glasses per week estimated as total grams of alcohol per week divided by 12 g of alcohol per glass.

Smoking status was categorized into never, past, and current smoking. Pack-years of smoking was calculated as the average number of cigarettes smoked per day/20 (i.e., 20 cigarettes in one pack) and multiplied by the number of years the subject smoked. Physical activity was estimated as number of hours per week of high-intensity physical activity (running, aerobic, or cycling for at least 30 minutes each time) and low-intensity physical activity (walking, gardening, snow clearing). We added up hours of low and high-intensity level physical exercise into one combined variable. We analyzed high, low, and the combined activity variables separately, but only present results for the combined low- and high-activity variable.

Weight and height were self-reported and are not as reliable as measured weight and height. Body mass index (BMI) was defined as weight (kg)/height (m²). Menopausal status was defined as premenopausal if a woman reported to still have a regular menstrual period, perimenopausal if she reported irregular periods, and postmenopausal if she reported menstruation had stopped or being on menopausal hormone therapy.

We considered the following variables as potential confounders: body mass index (BMI; ≤22, 23–25, 26–28, >28 kg/m²), education (no formal education/primary school, high school, bachelor/master), age at menarche (9–12, 13, 14, 15–18 years), number of pregnancies lasting at least 6 months (never, 1, 2, 3, ≥4), menopausal status (pre-, peri-, postmenopausal), and, where appropriate, alcohol intake (never drinkers, 1, 2, 3–4, and 5+ glasses per week), high intensity physical activity (no high intensity, 1, 2–3, 4+ hours per week), and smoking habits (never, past, and current).

If a variable was missing on all the questionnaires a woman had completed, we excluded her from all analyses. Of the 6,471 breast cancer cases, we excluded the following due to missing information: BMI (n = 532), education (n = 135), age at menarche (n = 229), number of pregnancies (n = 164), menopausal status (n = 59), smoking (n = 62), alcohol (n = 154), and physical activity (n = 184). This left us with 4,952 breast cancer cases for analysis. Of the 339,714 remaining women in the cohort, before we selected controls, we excluded women with missing information on: BMI (n = 67,813), education (n = 8,362), age at menarche (n = 14,818), number of pregnancies (n = 8,771), menopausal status (n = 6,632), smoking (n = 6,381), alcohol (n = 12,878), and physical activity (n = 16,205). This left us with 197,854 women in the cohort. Of these, we matched five controls per case on year of birth and year of screening, which left us with 24,760 controls for analysis.

Conditional logistic regression models were fitted, to estimate ORs (with 95% confidence intervals; CI) as a measure of association between the exposures of interest and breast cancer (overall and by subtypes), adjusted for confounders.

We matched on date of birth and date of screening with ±3 years. Because of concern there could still be residual confounding by age, we conducted a sensitivity analysis where we controlled more tightly for age at screening using a continuous variable, but the results remained largely the same, and we therefore decided that the matching variable with the matched analyses were sufficient.

Trend tests were performed by fitting ordinal values corresponding to exposure categories and testing whether the slope coefficient differed from zero. Case–case analyses were conducted to test for heterogeneity between breast cancer subtypes taking the luminal-A–like subtype as the reference. We used likelihood ratio tests comparing the likelihood ratio of the case–case model with confounders only to that of the same model also including the exposure variable.

We also assessed potential effect modification by BMI (<25, ≥25 kg/m²), parity (nulliparous, parous), smoking status (never, past, and current smokers), and estrogen and progesterone therapy (never hormone therapy users, current estrogen, and progesterone therapy users).

All analyses were performed using STATA (Stata Statistical Software: Release 14, StataCorp.). We considered a two-sided P value less than 0.05 as being statistically significant.

Both current and past alcohol consumption were positively associated with overall breast cancer risk (P_trend < 0.03 for all), with women who reported drinking 6+ glasses/week currently and at ages 40–49 and 30–39 years having, respectively, a 1.26 (95% CI, 1.09–1.45; Table 1), 1.34 (1.12–1.60), and 1.31 (1.06–1.62; Supplementary Table S1) increased risk relative to never drinkers (Table 1). Type of alcohol (beer, liquor, wine) was not independently associated with breast cancer overall (Table 1). Smoking status was positively associated with breast cancer overall (P_trend = 0.006), with current smokers having a 13% increased risk (OR = 1.13; 95% CI, 1.03–1.23) relative to never smokers (Table 1). The risk of developing a breast cancer increased progressively with increasing number of cigarettes smoked at the time of questionnaire completion (P_trend = 0.01; Table 1) and at ages 40–49 years and 30–39 years (P_trend < 0.0001) with women who smoked 20+ cigarettes per day having a 31%–41% increased risk relative to never-smokers (Supplementary Table S1). Positive trends in the odds of breast cancer with smoking intensity were also present when analyses were restricted to never drinkers (OR = 1.28; 95% CI, 0.87–1.90; current vs. never smokers). Current physical activity (low and high intensity combined) was negatively associated with overall breast cancer (P_trend = 0.01), with women who reported 6+ hours of physical activity per week having a 16% lower risk of developing breast cancer (OR = 0.84; 95% CI, 0.70–1.02) than those who reported never exercising (Table 1). Height and BMI were both positively associated with overall breast cancer risk (P_trend < 0.0001 for both; Table 1).

Analyses by breast cancer subtypes showed that current (Table 2) and past alcohol intake (Supplementary Table S2) were both positively associated with luminal A–like breast cancer risk (P_trend < 0.05 for all), with women who reported drinking 3+ glasses/week having an approximately 15% increased risk relative to those who reported drinking only 0–1 glasses/week. Alcohol intake (current; Table 2) and (at 30–39 years; Supplementary Table S2) was not associated with risk of luminal B–like HER2-negative disease (P_{heterogeneity} = 0.02). In contrast, current alcohol intake (P_trend = 0.01) and, to a lesser extent, intake at ages 30–39 (P_trend = 0.04) were inversely associated with risk of HER2-positive subtype albeit with no clear evidence of heterogeneity of effects with the luminal A–like subtype (Table 2). Alcohol intake was not associated with risk of triple-negative disease.

Relative to never smoking, current smoking was positively associated with luminal A–like breast cancer (OR = 1.18; 95% CI, 1.05–1.32). Women who reported smoking 10+ cigarettes/day, currently (Table 2) or in their 30s or 40s (Supplementary Table S2), were at a 19%–27% increased risk of luminal A–like subtype compared with never smokers. Risk of luminal B–like HER2-negative subtype was also positively associated with intensity of smoking (Table 2). Smoking was not associated with risk of any other breast cancer subtype.

Only current physical activity (low- and high-intensity combined) was inversely associated with risk of luminal A–like breast cancer (P_trend = 0.01), with women who reported currently exercising 4+ hours/week having a 15% lower risk (OR = 0.85; 95% CI, 0.75–0.97) than those who reported 0–1 hour/week (Table 2). Current physical activity was not associated with a decreased risk of luminal B–like HER2-negative breast cancer (P_{heterogeneity} = 0.08) or any of the other breast cancer subtypes.

We conducted a sensitivity analysis to determine whether the cases with no subtype information differed from those with subtype luminal A-like. We found no case–case differences for alcohol, smoking, and physical activity between luminal A–like cases and cases with no subtype information (not significant P_{heterogeneity}; Supplementary Table S3).

The results of a sensitivity analysis including luminal B–like HER2-negative in the luminal A-like subtype remained largely the same as the results when these subtypes were divided (Supplementary Table S4).

In this large nested case–control study, we found that alcohol was associated with an increased risk of luminal A–like breast cancer, smoking was associated with an increased risk of luminal A–like and luminal B–like HER2-negative breast cancer, and physical activity associated with a decreased risk of luminal A–like breast cancer. These modifiable risk factors were not associated with the other subtypes.

In our study, we defined luminal A–like slightly differentially than a number of other studies. Often other studies have included our luminal B–like HER2-negative subtype into the luminal A–like subtype. However, our results suggested that alcohol and physical activity were differently associated with luminal A–like and luminal B–like HER2-negative subtypes, suggesting that these subtypes have different etiology and should not be included into one luminal A–like subtype.

Several studies have investigated the association between alcohol and subtypes of breast cancer (7, 9, 13–16, 42). One study had information on ER only (15), two studies on ER and PR (9, 13), and four studies on ER, PR, and HER2 (7, 14, 16, 42). The pooled analysis from the Nurses' Health Study (NHS) reported that alcohol consumption was associated with an increased risk of both ER⁺ and ER⁻ breast cancer (15). Although this was a well powered study (more than 20,000 ER⁺ cases), no results were reported for more detailed subtypes. Other studies with both ER and PR information and moderate sizes (2,500–2,800 cases) have found that alcohol was positively associated with hormone receptor–positive breast cancer, and not with and ER⁻PR⁻ tumors (9, 13, 14).

These inconsistencies are also found in studies with HER2 information. The NHS (2,760 cases) found that alcohol was associated with an increased risk for luminal A and HER2-positive breast cancers (16). On the other hand, a Spanish case–control study with 1,766 cases reported a positive association between alcohol and all breast cancer subtypes, except HER2-positive subtype (7). Furthermore, the large European Prospective Investigation into Cancer and Nutrition (EPIC) cohort (11,576 cases) reported a positive association between alcohol consumption and ER⁺PR⁺, ER⁻PR⁻, HER2⁻ and ER⁻PR⁻HER2⁻ breast cancers (42), but again, no association between alcohol and HER2-positive breast cancer. Overall, there is consistency in the results that alcohol is associated with luminal A–like breast cancer. This is also consistent with our finding. We found no significant positive association with HER2-positive or triple-negative breast cancer, similarly to some, but not all of the previous studies. The differences in sample size do not seem to explain the alcohol discrepancies.

When we considered type of alcohol separately and adjusted for the other types of alcohol, intake of beer, liquor, and wine were not associated with any of the subtypes. These results are not consistent with the findings in the Women's Health Initiative study, which reported that wine and liquor were positively associated with ER⁺ breast cancer (14), possibly because they did not adjust for other types of alcohol.

It has been suggested that the biological mechanisms of alcohol on breast cancer can be both hormonal and nonhormonal. The nonhormonal explanation is that alcohol is metabolized to acetaldehyde that can induce DNA damage (43). Acetaldehyde may also be broken down to reactive oxygen species (ROS), which could cause breast malignancies via DNA mutation, base deletion, and single and double strand breaks (43). The hormonal explanation is that alcohol intake is associated with increased levels of estrogen among both pre- and postmenopausal women (44–48), and alcohol has also shown to influence the estrogen receptor (30, 34, 49) and may affect the breast tissue's sensitivity to estrogen and enhance growth in ER-positive tumors (32). Our findings alone would be compatible with the hormonal explanation, but not with the finding in the EPIC study where alcohol was associated also with ER⁻PR⁻ tumors.

Several studies with information on ER, PR, and HER2, have looked at the association between smoking and subtypes of breast cancer (14, 19, 20, 22, 24). One of these studies was a small study (224 cases; ref. 19), another included 900 cases (20) and three studies contained from 1,800 to 2,500 cases (14, 22, 24). Despite the differences in sample size, all these studies reported that smoking was associated with an increased risk of ER⁺ and luminal-like breast cancers, and no association with triple-negative and basal-like breast cancers. This is consistent with our finding of a positive association between smoking and luminal A–like and luminal B–like breast cancer and no association with triple-negative tumors. The fact that we separated the luminal B–like HER2-negative from luminal A–like subtype, could be one explanation of our finding of a positive association between smoking with both luminal A–like and luminal B–like HER2-negative breast cancer.

The presence of chemicals generated by smoking has been detected in mammary gland tissue of both healthy women and breast cancer patients (50–52). In vitro studies suggest that nicotine (31), benzo(a)pyrene (53), 1-methylanthracene (54), and penanthrene (55) may provoke alterations in epigenetic and transcriptional mechanisms which regulates the expression of genes that are involved in transformation of healthy epithelial cells, proliferation, and metastasis of tumor epithelial cells into the mammary gland (26). There are carcinogens in tobacco smoke such as polycyclic aromatic hydrocarbons (PAH), aromatic amines, and nitrosoamines (20), and these carcinogens can have both estrogenic (56) and antiestrogenic effects (57, 58). The estrogenic effect of these carcinogens can explain the current study finding of a positive association between smoking and luminal A–like and luminal B–like HER2 negative breast cancer.

The association between physical activity and breast cancer subtypes have been investigated in several studies (26–32, 35–37). Some of these studies had information on ER only, ER and PR (27, 30–32, 37), and others had information on ER, PR, and HER2 (26, 28, 29, 35, 36). Of all these studies, two of them had fewer than 700 cases (29, 35) and six studies included from 1,921 to 5,882 cases (26–28, 31, 32, 36). The majority of these studies found that physical activity was associated with a decreased risk of ER⁺, ER⁺PR⁺ and luminal A–like subtype, and found no association with ER⁻, ER⁻PR⁻ and triple-negative breast cancers. Consistent with these findings, our study found that physical activity was associated with a decreased risk of luminal A–like breast cancer, and no association with the other subtypes. However, a review of 47 studies reported a greater risk reduction of physical activity on ER⁻PR⁻ tumors than on ER⁺PR⁺ tumors (30), but the review did not say whether the results for ER⁻PR⁻ tumors were statistically significant. In addition, this latter review only included studies with information on ER and PR. The California Teachers Study found that long-term and baseline strenuous physical activities were associated with a reduction only in triple-negative breast cancer, but the association did not reach statistical significance (28).

Physical activity improves insulin sensitivity (59) and decreases estradiol bioavailability by increasing sex hormone–binding globulin (SHBG) levels (60). Furthermore, nonhormonal mechanisms involved in physical activity are decreased inflammation (59), interference with the glycolytic switch (61), decreased DNA damage, or immunomodulation (59). We did not find a statistically significant interaction between estrogen and progesterone therapy and physical activity, but we observed a protective effect of physical activity only in never users, suggesting that biological mechanisms of physical activity on breast cancer works through hormones. The hormonal explanation is also consistent with our finding, and that of others, that physical activity appears to protect predominantly against luminal A–like cancer.

Strengths of this study include its population-based design, the large size, being one of the largest single studies on breast cancer subtypes conducted so far, and the availability of prospectively collected detailed information on many risk factors for breast cancer.

Another strength is that we did not combine luminal A–like subtype with the luminal B–like HER2 negative as many other studies have done. Our results suggested that alcohol and physical activity were associated differently with these two subtypes, suggesting that these should be treated as two different subtypes.

Although this study is one of the largest to date on breast cancer subtypes, there was still limited power for the rare breast cancer subtypes. Another weakness was the lack of molecular expression data. This may have obscured differences between the subtypes.

Women who attend screening might be more health conscious and have a healthier lifestyle than women who do not attend. This could have contributed to obliterate protective effects of physical activity and the increased risk of alcohol and smoking. At the same time, women who attend screening are more likely to have their breast cancers detected. The associations of alcohol, smoking, and physical activity with overall breast cancer risk were largely as expected. Furthermore, it is unlikely that any such bias would have differentially affected the subtype results.

In this large nested case–control study, alcohol and smoking were associated with an increased risk and physical activity with a decreased risk of breast cancer overall and with luminal A–like breast cancer. In contrast, none of these risk factors were associated with HER2-positive or triple-negative breast cancer.

No potential conflicts of interest were disclosed.

Conception and design: M. Ellingjord-Dale, S. Hofvind, I. dos-Santos-Silva, G. Ursin

Development of methodology: L. Vos, I. dos-Santos-Silva, G. Ursin

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M. Ellingjord-Dale

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M. Ellingjord-Dale, L. Vos, H.G. Russnes, S. Tretli, S. Hofvind, I. dos-Santos-Silva, G. Ursin

Writing, review, and/or revision of the manuscript: M. Ellingjord-Dale, K.V. Hjerkind, A. Hjartåker, H.G. Russnes, S. Tretli, S. Hofvind, I. dos-Santos-Silva, G. Ursin

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M. Ellingjord-Dale, L. Vos, S. Hofvind

Study supervision: G. Ursin