Alcohol, Smoking, and Risks of Breast Cancer Recurrence and Mortality among Women with Luminal, Triple-Negative, and HER2-Overexpressing Breast Cancer

In the United States, breast cancer is the most common type of cancer diagnosed among women. Given major improvements in breast cancer survival rates there are an estimated more than four million breast cancer survivors in the United States (1). Approximately 10%–20% of breast cancer survivors will experience a recurrence in the decade after diagnosis, and >90% of breast cancer–related deaths are due to metastatic disease (2–5). Fear of breast cancer recurrence is common among breast cancer survivors, many of whom may not be receiving survivor-specific guidance on health behaviors from health practitioners (6, 7). Although American Cancer Society guidelines for breast cancer survivors to avoid smoking are based on evidence from observational studies of breast cancer outcomes in relation to smoking status at diagnosis and randomized controlled trials of smoking cessation programs, there is no epidemiologic evidence informing guidelines on alcohol consumption (8). Furthermore, breast cancer is a heterogenous disease that can be divided into four molecular subtypes using joint hormone receptor (estrogen receptor, ER, and progesterone receptor, PR) and HER2 status: luminal A (ER⁺/HER2⁻), luminal B (ER⁺/HER2⁺), triple-negative (TN; ER⁻/PR⁻/HER2⁻), and HER2-overexpressing (H2E; ER⁻/HER2⁺; ref. 9). Luminal A is the most common subtype and has the best prognosis, followed by luminal B (9). TN disproportionately affects Black women and is associated with the worst prognosis, with no targeted treatment available (10). H2E breast cancer is another aggressive subtype, but targeted therapy (trastuzumab) is available (11).

The association of alcohol and smoking with breast cancer outcomes, overall and by molecular subtype is still not fully understood. Alcohol consumption may promote breast tumor growth, increase the potential of the tumor to metastasize, contribute to oxidative stress in breast tissue, and inhibit response to treatment (12). Several epidemiologic studies have identified increased risks of breast cancer outcomes associated with alcohol consumption (13–16); however, multiple studies have not observed an association (17–20) and others have observed decreased risks (21–25). Alcohol consumption is variably associated with risks of the different molecular subtypes of breast cancer (26–28), yet few studies have evaluated the association between alcohol and breast cancer outcomes by subtype (13, 15, 19, 29).

Poor breast cancer outcomes among smokers may occur due to effects on cell biology (30–32), response to treatment (30, 31), and treatment non-adherence resulting from toxicities (30, 31); however, the epidemiologic evidence of an association between smoking and breast cancer outcomes is conflicted (16, 33–38), To our knowledge, only two existing studies (39, 40) of smoking among women with less common molecular subtypes have been conducted, and they are limited by small sample sizes.

The purpose of this study is to investigate the relationship between smoking, alcohol, and breast cancer recurrence and mortality in a prospective, population-based cohort with an oversampling of the less common TN and H2E breast cancer subtypes.

Women ages 20 to 69 diagnosed with a first primary invasive breast cancer between June 1, 2004 and June 30, 2015 in the Seattle–Puget Sound region (King, Pierce, and Snohomish counties) were identified from the region's Surveillance, Epidemiology, and End Results (SEER) cancer registry. This study population has also been described in detail elsewhere (41, 42). To maximize enrollment of participants with less common molecular subtypes, eligible cases were all identified as TN and H2E cases, and a random sample of luminal (ER⁺) cases, frequency matched on age and year at diagnosis of TN/H2E cases combined. Eligibility criteria required complete information on tumor marker status. Vital status was not part of eligibility criteria, and a waiver of consent allowed for medical record review of deceased participants. 4,508 new cases were determined to be eligible, of which 2,882 were enrolled, for a response rate of 63.9%. In addition, the research team identified 994 eligible cases from prior studies with similar questionnaires who were diagnosed between June 1, 2004 and June 30, 2010, and otherwise met the eligibility criteria (43, 44) and included previously collected medical record and questionnaire data from 100% of these cases (95% of which consented for medical record review). In total, 3,876 cases were enrolled—2,153 luminal, 1,252 TN, and 471 H2E. Using a prospective cohort design, participants were followed forward in time for ascertainment of second breast cancer events and vital status. Written, informed consent was obtained for all study participants who were alive at study enrollment. A waiver of consent was obtained for deceased participants. This study conducted in accordance with U.S. Common Rule and was approved by the Fred Hutchinson Cancer Center Institutional Review Board.

Data on epidemiologic, demographic, and clinical factors at the time of initial study enrollment were collected using both structured interviewer-administered questionnaires and medical record reviews. Deceased participants only had data from medical record reviews, and some patients refused interview but consented to medical record review. Alcohol use and smoking status at diagnosis were collected for all participants, with variables defined primarily using data from structured interviewer-administered questionnaires, with data from medical record review used to supplement when questionnaire data were unavailable or incomplete. The majority of participants missing questionnaire data (95%) were deceased. To quantify the frequency and duration of alcohol use and smoking history, average drinks per week and total pack-years of smoking were ascertained. Post-diagnostic alcohol use and smoking status were collected using mailed questionnaires for the subset of participants for whom additional follow-up was performed to assess concordance of pre- and post-diagnostic measures. Other covariates (including body mass index, hypertension, and diabetes) were defined prioritizing data from medical records, with data from structured interviewer-administered questionnaires used to supplement when questionnaire data were unavailable or incomplete. Subtypes of breast cancer were defined by joint ER/PR/HER2 status using clinical data abstracted from patient medical records. HER2/neu, as abstracted from medical records, was determined by the combination of IHC stains and FISH. IHC was used to classify HER2 as positive or negative, but if IHC results were equivocal, FISH test was used to clarify the result.

Data on breast cancer recurrence and second primary breast cancers, including anatomical site of recurrence (locoregional, distant, lung, liver, bone, brain) were ascertained through medical record review only. Data on non-breast second primary cancers were not collected. Vital status and cause of death were obtained through linkages to SEER through June 15, 2022. SEER provides a variable for cancer-specific cause of death that indicates that a person died as a result of their cancer (45). To minimize misclassification in the creation of this variable, SEER uses an algorithm that considers causes of death, tumor sequence, comorbidities, and site of the primary cancer diagnosis.

To ascertain long-term recurrence (5–16 years after diagnosis), we identified a subset of 1,600 participants (41.3% of the original sample) alive and without a recurrence at last follow-up who had at least one previous medical record reviewed from area institutions that allow electronic access to medical records. In total, 682 participants (42.6%) completed post-diagnostic questionnaires on alcohol use and smoking status after their diagnosis and 705 (44.1%) medical record reviews for ascertainment of recurrence were completed.

We assessed the distribution of patient characteristics and known breast cancer risk factors for participants overall and by molecular subtype. Multivariable-adjusted Cox proportional hazards regression were fit to estimate hazard ratios (HR) and associated 95% confidence intervals (CI) separately for each molecular subtype and for cases overall. All models included mutual adjustment for alcohol or smoking at diagnosis and adjusted for the following categorical variables measured at diagnosis: Age (20–39, 40–49, 50–59, 60–69 years), year of diagnosis (2004–2006, 2007–2008, 2009–2011, 2012–2015), stage, history of hypertension, history of diabetes, race/ethnicity (non-Hispanic White, Hispanic White, Black, and other), and body mass index (<25, 25≤30, 30+). Adjustment variables were selected a priori using a directed acyclic graph. Because of difference in baseline hazard of recurrence and mortality among molecular subtypes, subtype was included as a stratification variable, allowing the baseline hazard to vary for each subtype for analyses across subtypes. Trend tests for the association of drinks per week among current alcohol users, and total pack-years smoked among participants who had ever smoked were performed using Wald tests. Models with an interaction term for alcohol use and smoking status were also fit to assess whether there were joint effects for alcohol and smoking. Because of the small sample of luminal B cases, we combined luminal A and B cases into one ER-positive luminal group for analyses to maximize power. To account for selection effects resulting from the under-sampling of luminal cases into the overall cohort, inverse probability weighting was applied in analyses examining risk for all subtypes combined (46, 47). For subtype-specific analyses, multivariable-adjusted Cox proportional hazards regression models were fit separately for each molecular subtype. Missing values of exposures and other covariates were imputed using multiple imputations by chained equations (48). Variables were not imputed if missingness was greater than 30%. Outcome variables were included in the models used to impute other variables but were not imputed by themselves; thus, analyses were restricted to those with complete information on outcomes. Effect modification by menopausal status and year of diagnosis was assessed with a Wald test after imputation, with two-sided P values less than 0.05 considered statistically significant. All analyses were conducted using R software in RStudio.

Survival time was measured as the time from breast cancer diagnosis until first recurrence or censoring. Cases were censored at last follow-up if they had not experienced a recurrence. Patients were also censored if they experienced a second primary breast cancer. We evaluated any recurrence and characterized recurrence by anatomical site (locoregional, distant, lung, liver, bone, and brain). For analyses with anatomical site of recurrence as the event of interest, multivariable-adjusted Cox proportional hazards regression models were fit for each site-specific recurrence accounting for competing-risks using cause-specific hazard models, with molecular subtype included as a stratifying variable. Analyses were restricted to patients without stage IV disease at diagnosis and who had medical record reviews. Patients with a recurrence that occurred within 6 months of their breast cancer diagnosis were excluded from recurrence analyses, as these events are less likely to be related to second breast cancer events, but rather resulting from primary resistance of the initial primary breast cancer.

Breast cancer–specific and all-cause mortality were evaluated, with survival time defined as the time from breast cancer diagnosis until death or censoring. Participants of all stages were included in mortality analyses. Participants still alive at last known follow-up were censored. In breast cancer–specific mortality analyses, cause-specific Cox proportional hazards models were used, and participants who died of other causes were censored at the date of their death.

Multiple sensitivity analyses were conducted. First, we re-fit models with luminal A and B cases categorized separately, to evaluate whether results differed when luminal A and B cases are not combined. Because cases with certain exposures may be less-likely to complete guideline-concordant treatment, we also performed analyses where cases who did not receive definitive local treatment (defined as either breast-conserving surgery plus radiotherapy or mastectomy, with or without radiotherapy) were excluded. Finally, to take into account the time-varying hazard of breast cancer recurrence by subtype (49) and to evaluate how exposures of interest may impact risk of recurrence at different time periods after diagnosis, we compared hazards for different time periods following diagnosis using time-dependent coefficients (50).

The data generated in this study are available upon request from the corresponding author.

Distribution of participant characteristics is presented in Table 1. Participants with luminal breast cancer were more likely to be diagnosed at earlier stages, and be non-Hispanic White compared with other participants. TN cases were more likely to have a higher BMI at diagnosis, be Black, and have hypertension and diabetes at diagnosis than other subtypes, and H2E cases were more likely to be diagnosed at stage III or IV.

In this prospective cohort, 467 breast cancer recurrences were observed, with a median follow-up time of 4.27 years (range, 0.5–16.54 years) and a median time to recurrence of 1.87 years (range, 0.5–16.31 years; Table 2). Of the observed recurrences, 66 (14%) occurred five or more years after initial breast cancer diagnosis. 933 deaths (662 due to breast cancer) were observed, with a median follow-up time of 10.75 years (max, 17.83 years) and a median time to breast cancer–specific death of 3.08 years (max: 14.83) and to death due to any cause of 3.68 years (max, 17). Women with luminal breast cancer were more likely than other subtypes to experience a first disease recurrence five or more years after their initial diagnosis (Fig. 1).

Current alcohol use at diagnosis, relative to not current use, was not associated with an increased risk of breast cancer recurrence, overall (HR, 0.98; 95% CI, 0.74–1.28) or by molecular subtype (Table 3). However, current alcohol use of four or more drinks per week was associated with a decreased risk of recurrence among women with TN breast cancer (HR, 0.71; 95% CI, 0.51–0.98). When examining risk of locoregional and distant recurrence, no statistically significant associations were observed for current consumption of alcohol among all cases or within any subtypes (Supplementary Table S1). Also, among TN cases, current alcohol consumption of four or more drinks per week was associated with a lower risk of breast cancer–specific mortality (HR, 0.73; 95% CI, 0.55–0.97) and all-cause mortality (HR, 0.68; 95% CI, 0.53–0.86; Table 3).

Relative to never smoking, a history of ever smoking was associated with a 33% increased risk (95% CI, 1.01–1.74) of breast cancer recurrence only among TN cases, and current smoking at diagnosis was associated with a 59% greater risk (95% CI, 1.07–2.35; Table 4). Current smoking was associated with an increased risk of distant recurrence for breast cancer overall (HR, 1.53; 95% CI, 1.02–2.30; Supplementary Table S2). A history of ever smoking and current smoking at diagnosis was associated with over 50% and 80% greater risks, respectively, of recurrence to bone among luminal cases and cases overall (Supplementary Table S2). Compared with never smoking, ever smoking was associated with 30% to 50% greater risks of breast cancer–specific death among cases overall, luminal cases, and TN cases and 53% to 61% greater risks of death from any cause across subtypes, with higher risks observed for consumption of 10 or more pack-years (Table 4). Current smoking was associated with 50% to 131% greater risks for breast cancer–specific mortality and 122% to 158% greater risks for all-cause mortality across subtypes.

When the time scale was stratified, using time-dependent coefficients, estimates were not statistically significant. Estimates for breast cancer–specific mortality did not greatly differ by time since diagnosis, with one exception; there was a stronger association observed for current smoking status and early breast cancer–related death (<5 years) for luminal cases and late breast cancer–specific death (≥5 years) for TN cases. In models allowing for an interaction between alcohol use and smoking status, there was no evidence for a joint effect on any outcome.

In this prospective cohort study with a large sample of TN and H2E breast cancers, we examined the risks of breast cancer recurrence and breast cancer–specific mortality associated with alcohol use and smoking status at diagnosis. We found no clear association between current alcohol use at diagnosis and breast cancer recurrence or mortality in cases overall; however, we did observe a decreased risk of breast cancer recurrence and mortality among TN cases who currently consumed four or more drinks per week at diagnosis. Similarly, other studies have failed to identify an association between alcohol use and breast cancer recurrence or breast cancer–specific mortality (17–20, 25). Four other studies (21–23, 25) and one meta-analysis (24) found inverse associations between alcohol consumption and breast cancer outcomes, and the present study builds on these findings by including larger numbers of women with less-common molecular subtypes. To our knowledge, only one other study (29) has investigated the relationship between alcohol consumption and outcomes among women with TN breast cancer, with null findings for overall and disease-free survival; however, the sample size of 197 may not have provided adequate power to detect decreases in risk and did not allow for analyses by quantity of alcohol consumed as we performed. Our findings of an absence of a relationship between alcohol use and breast cancer outcomes are in contrast with other studies that have reported increased risks of breast cancer recurrence associated with pre-diagnostic alcohol consumption (13, 14), and increased risks of both recurrence (15, 16, 19) and breast cancer–specific mortality (15) associated with post-diagnostic drinking. However, these studies did not collect information on HER2 expression, and thus could not examine outcomes by molecular subtype.

Regular alcohol use may activate the immune system, which could in turn impact the success of chemotherapy (51). One study has shown light alcohol consumption to suppress breast cancer progression in an animal model (52), but the effects we observed were at a higher consumption level. The discrepancies between some of our results and findings from other studies may also be due, in part, to differences in how alcohol use was measured between the present and other studies. Most of the studies discussed above used questionnaires to ascertain alcohol consumption [aside from one study (ref. 29) that used data from medical records], whereas in the present study we prioritized data from structured, interviewer-administered questionnaires and used data from medical records when questionnaire data were missing. Furthermore, because deceased participants were included in this study through a waiver of consent for medical record review, 95% of those missing questionnaire data were deceased, which could have resulted in differential misclassification of alcohol consumption. Although concordance between medical records and questionnaires for current alcohol use among those who had data from both sources was 78%, concordance for drinks per week was only 51%. However, results from sensitivity analyses restricted to those with data from questionnaires did not differ greatly, with the inverse association observed for alcohol use among TN cases persisting in magnitude, indicating that misclassification is not fully responsible for the observed findings. In addition, we did not have information on socioeconomic status or other lifestyle characteristics, such as diet or physical activity, which may confound the relationship between alcohol use and breast cancer outcomes. Thus, residual confounding from these variables may explain the decreased risk of recurrence associated with alcohol use we observed among women with TN breast cancer.

We found that history of ever or current smoking at diagnosis was associated with an increased risk of breast cancer recurrence among TN cases and with an increased risk of breast cancer–specific mortality across all subtypes. In addition, we identified a relationship between smoking and risk of recurrence to bone among luminal cases and cases overall. The associations between smoking and breast cancer outcomes we identified are consistent with findings from some (16, 37–40), but not all (16, 33) previous studies; however, these previous studies had lower sample sizes of women with ER⁻ or HER⁺ tumors. In regard to findings by specific molecular subtype, one prospective cohort of postmenopausal breast cancer cases in Germany also found greater risks of adverse breast cancer outcomes associated with smoking at the time of diagnosis for TN and luminal A-like tumors (39). In contrast with our findings, this study (39) also found an over 3-fold greater risk of recurrence for H2E breast cancer associated with smoking; however, sample sizes for TN (N = 331) and H2E (N = 177) cases were smaller than those of the present study. Another majority postmenopausal study (40) found poorer disease-free survival among 236 women with trastuzumab-treated HER2⁺ breast cancer who smoked before diagnosis; however, only 79 had ER⁻ tumors, which was much smaller than our sample of women with H2E breast cancer. In addition, both of these studies included more postmenopausal participants than ours (51.5% of our sample were postmenopausal) so results may not be directly comparable, but we did not find evidence of effect modification by menopausal status in primary analyses. A retrospective study of 131 women with TN breast cancer undergoing breast reconstruction did not identify an association between smoking and breast cancer outcomes; however, the sample size was small (53).

By altering cell signaling pathways, cigarette smoke can increase proliferation and tumorigenesis, and can increase cancer cells’ ability to evade apoptosis (30, 31). Smoking may also inhibit response to and increase toxicity from chemotherapy and radiotherapy, making these therapies less efficacious (30, 54). We observed associations between smoking and increased risk of recurrence only for the TN subtype, which is not able to be treated with hormone therapy or HER2-targeted treatment due to its lack of expression of ER, PR, or HER2 receptors. As chemotherapy is the main choice for systemic treatment of TN breast cancer, if past or current smoking inhibits the response to chemotherapy this could allow breast cancer cells to metastasize in the absence of another therapeutic agent. Several genetic and metabolic factors may have a role in the pathology of bone metastases in women with breast cancer, and cigarette smoke may interact with these factors to increase the risk of breast cancer cells homing to and invading the bone (55). In addition, as smoking is known to increase the risk of osteoporosis (56) it may degrade the bone structure in such a way that allows for a greater likelihood of metastasis; however, this risk may be mitigated by use of bone stabilizing agents after chemotherapy. Similar to the discussion of findings for alcohol consumption, there was a potential for differential misclassification of smoking status due to reliance on medical records for participants who were deceased at study enrollment; however, concordance between medical records and questionnaires for those who had data from both sources was even greater for smoking status (93%) and pack-years smoked (79%) than for measures of alcohol use. Again, results from sensitivity analyses restricted to those with data from questionnaires did not differ greatly from findings we present here. Unmeasured confounders of the relationship between smoking and breast cancer outcomes included socioeconomic status and other lifestyle characteristics, such as diet or physical activity, and residual confounding from these variables could have resulted in risk estimates of larger magnitude. However, one study of ER⁺ breast cancer (16) that adjusted for physical activity presented similar risk estimates to results among our cases with ER⁺ disease.

A major strength of this study is the large, population-based sample of women with the less common TN and H2E breast cancer subtypes, relative to previous studies, allowing findings to be generalizable to more than just luminal breast cancer. In addition, with just up to 16 years of follow-up for recurrence and 17 years of follow-up for mortality for some participants, this study provides a better understanding of how these exposures may impact long-term outcomes. However, the present study is not without limitations. First, we used measures of alcohol consumption and smoking status for the period at or just before initial breast cancer diagnosis, when post-diagnostic measurements may be of greater interest. However, among the subset of cases who completed questionnaires at diagnosis and post-diagnostic questionnaires, 90% reported that they drank alcohol, and 91% had the same smoking status, at the time of their diagnosis and at follow-up. Biologically, metastasis has been hypothesized to be a relatively early event in the natural history of cancer, as dissemination of metastatic cancer cells can occur before the primary tumor is even discovered (57). So, the strength of the availability of data on measures before initial breast cancer diagnosis is that these measures allow us to examine their effect on breast cancer recurrence through mechanisms that occur early on in the disease course. In addition, pre-diagnostic measures are not subject to reverse causation, as disease severity, treatment, and undiagnosed recurrence could impact post-diagnostic patterns of alcohol consumption and smoking. Second, because of the small sample of luminal B cases, we were not able to evaluate these cases separately and instead grouped these cases with luminal A cases in our main analyses. We did perform sensitivity analyses with luminal A and B cases categorized separately, and results did not greatly differ with the addition of luminal B cases. Adjustment for confounding variables was pre-specified using a directed acyclic graph, and we chose not to include treatment characteristics as adjustment variables in our models as they are on the causal path between exposure variables and outcomes and would thus be mediators instead of confounders. However, in a sensitivity analysis, we restricted the sample to those who received definitive local treatment (breast conserving surgery and radiotherapy or total mastectomy) and our results did not change. Furthermore, adjustment for definitive local treatment, chemotherapy, hormonal therapy (for luminal and overall only), and targeted treatment (for luminal, H2E, and overall only) did not greatly change risk estimates so treatment variables were not added to final models. In addition, this study recruited participants over 11 years (2004–2015), and although study methods did not greatly differ across this time, participants diagnosed in later years were more likely to never smoke, less likely to have received definitive local treatment, less likely to have received hormonal treatment, and more likely to be diagnosed with stage IV disease. Trastuzumab was also FDA approved as adjuvant treatment for HER2⁺ breast cancer in 2006, reducing mortality from early-stage, HER2⁺ breast cancer by a third (58). To account for these differences, we adjusted for year of diagnosis in all models. In addition, we found no evidence of effect modification by year of diagnosis (P > 0.05) and thus did not stratify results by year of diagnosis. Finally, extended follow-up for recurrence was only available for a subset of participants (N = 705) who were slightly more likely to have luminal breast cancer, be diagnosed more recently, and be diagnosed at earlier stages. However, the risk of recurrence for luminal breast cancer has been shown to persist longer into follow-up than for other subtypes, so longer follow-up in this subset is reasonable (49, 59).

Overall, the findings of this study indicate that patients with breast cancer with a history of smoking at diagnosis are at a greater risk of adverse outcomes, including recurrence to the bone. This relationship underscores the need to prioritize smoking cessation among women diagnosed with breast cancer and may indicate a role for adjuvant treatment with bisphosphonates to prevent bone metastases in luminal breast cancer. Future studies are needed to confirm the lack of association between alcohol use and breast cancer outcomes in the setting of less-common molecular subtypes and current treatment guidelines.

N.C. Lorona reports grants from NCI during the conduct of the study. M. Othus reports personal fees from BMS, Grifols, Biosight, Merck, and Glycomimetics outside the submitted work. K.E. Malone reports grants from NIH/NCI during the conduct of the study. M.T.C. Tang reports grants from Department of Defense Breast Cancer Research Program and NCI during the conduct of the study. No disclosures were reported by the other authors.

N.C. Loroña: Conceptualization, data curation, software, formal analysis, methodology, writing–original draft, writing–review and editing. M. Othus: Methodology, writing–review and editing. K.E. Malone: Methodology, writing–review and editing. H.M. Linden: Methodology, writing–review and editing. M.T.C. Tang: Writing–review and editing. C.I. Li: Conceptualization, supervision, funding acquisition, methodology, writing–review and editing.

This work was supported by the NCI [grant numbers T32CA009168 (to N.C. Loroña) and 261201000029C (to C.I. Li)] and the Department of Defense Breast Cancer Research Program (grant number: BC112721; to C.I. Li).

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.