Factors associated with and patterns of alcohol intake in late survivorship for breast cancer survivors.

Introduction
Breast cancer (BC) is the most common cancer diagnosis among women in the United States [1], and in 2020, the worldwide BC incidence surpassed lung cancer for the first time [2]. In recent decades, the overall incidence rates of BC in females have increased, specifically by 1% annually since 2012 [1]. With advances in systemic and local treatments as well as earlier detection through improved screening, death rates due to BC have decreased by 44% since 1989, preventing over 500,000 estimated deaths [1]. Given high rates of survival, with the 5-year survival rate for BC exceeding 90%, it is anticipated that an ever-increasing population will be living with or in remission from BC [1]. Thus, evaluating and addressing factors which negatively influence BC outcomes is of the utmost importance.
Alcohol is a known carcinogen and has been linked to an increased BC incidence [3–7]. Patients with a history of certain cancers, including BC, significantly benefit from healthy lifestyle choices [8]. Approximately, 30% of BC cases can be attributed to modifiable lifestyle factors, including alcohol intake [6]. Recurrence and mortality have been found to be impacted by both pre- and post-diagnosis alcohol consumption levels, with studies suggesting that higher alcohol consumption is associated with an increased risk of overall mortality, and post-diagnosis alcohol intake is positively associated with breast cancer recurrence [4]. Some studies also suggest a higher risk of recurrence in postmenopausal but not premenopausal women who regularly consume ≥ 6 g of alcohol daily post-diagnosis (HR, 1.19; 95% CI 1.01–1.40) [9].
Most of the available evidence has focused on the relationship between alcohol intake and BC incidence, and the results are relatively consistent [10]. Less has been published on post-diagnosis alcohol intake and the risk of BC recurrence. Additionally, there are little or no studies assessing alcohol intake in late BC survivorship (often defined as the period from 2 to 5 years post-diagnosis) [11–13]. Multiple studies show that, following a BC diagnosis, patients are likely to alter their dietary habits and alcohol intake (i.e., eating a more balanced diet and/or reducing alcohol intake) [14, 15]. While cancer survivorship guidelines suggest limiting alcohol intake, if not avoiding it altogether [16], some cancer survivors continue to consume alcohol, and may even increase their intake post BC diagnosis. Studies suggest roughly 75% of cancer survivors drink alcohol post-diagnosis, with many partaking in risky consumption patterns (consuming > 2 drinks per day, binge drinking, i.e., consuming ≥ 6 drinks on a single occasion, hazardous drinking, i.e., AUDIT-C scores ≥ 3)[14], [17, 18]. Many studies do not monitor long-term alcohol intake in BC survivors, which is critical as these patients are now living longer than ever before.
Our group previously studied demographic, lifestyle, and clinical characteristics associated with changes in alcohol intake in a cohort of BC survivors in early survivorship, i.e., in the year following BC diagnosis [19]. However, alcohol use is not static across survivorship; it is shaped by evolving clinical, psychological, and social factors. Our current study therefore aims to assess changes in and associations with alcohol intake in the same cohort of BC survivors later in survivorship, i.e., 4 years following BC diagnosis.

Methods
Individuals aged 18 or older who were diagnosed with BC within the past year and had at least one clinical visit at the Mayo Clinic in Rochester, Minnesota, were prospectively enrolled in the Mayo Clinic Breast Disease Registry (MCBDR) upon providing informed consent. This study represents a secondary analysis of data collected prospectively through the MCBDR. Participants diagnosed with BC between 2014 and 2022 were asked to provide information regarding their alcohol intake on the initial (baseline) survey and again at approximately 4 years after diagnosis.
Individuals who completed the alcohol intake question on both the baseline and 4-year follow-up surveys were included in the analysis. Exclusion criteria included a diagnosis of stage IV BC, a documented BC recurrence prior to the completion of the Year-4 survey, or a baseline survey submitted more than one year after the initial diagnosis.
The baseline and Year-4 surveys assessed alcohol intake by a single question asking participants to estimate their average number of alcoholic drinks consumed per week over the past 10 years, defining a standard drink as 5 oz of wine, 12 oz of beer, or 1 oz of liquor. This measure was not derived from a validated structured interview such as the Timeline Follow-Back, and no visual aids were used. Alcohol use response options were “none,” “1–4,” “5–9”, “10–14,” “15–19,” “20–29,” “30–39,” “ > 39,” and “I don’t know.” For this analysis, alcohol use data were categorized as: none, low (1–4 drinks), moderate (5–14 drinks), and high (15 or more drinks).
Participants provided clinical data and additional demographic information regarding their education level, the presence or absence of financial strain (e.g., difficulty paying bills or limited disposable income), their smoking behavior (never, former, current), exercise levels, menopausal status, and PROMIS-10 health scores. For exercise levels, participants reported the number of minutes per day and days per week of engagement in mild, moderate, and strenuous exercise. Mild-intensity activity was defined as requiring “minimal effort,” moderate-intensity activity as “not exhausting,” and strenuous activity as “heart beats rapidly.” The PROMIS Global-10 instrument consists of 10 items evaluating overall health, well-being, and distress, yielding scores for global physical health (GPH) and global mental health (GMH) [20]. These scores are presented as T-scores derived from raw totals, where a T-score of 50 represents the average of the U.S. general population, with a standard deviation of 10. Higher T-scores indicate better physical or mental health. T-scores were categorized based on approximate quartiles among all participants returning baseline questionnaires. Data were abstracted by trained nurses from electronic medical records—including patient race, gender, age at diagnosis, cancer stage and features, BRCA mutation status, and post-diagnosis treatments (such as surgery, radiation, chemotherapy, and/or endocrine therapy)—and incorporated into the analysis.

Statistical methods
Data were summarized using frequencies and percentages for categorical variables and means and standard deviations for continuous variables. Associations of participant attributes with return of the Year-4 questionnaire, among all patients returning the baseline questionnaire, were assessed using chi-square tests.
Among participants returning the Year-4 questionnaire, univariable cross-sectional associations of Year-4 alcohol intake with demographic and clinical characteristics were carried out using Monte Carlo-based Fisher exact tests. In brief, for a given comparison, a cross-tabulated table was created using the observed data, and the Fisher test statistic was calculated. Following this, 10,000 random samples of cross-tabulated tables were generated, conditional on the same total sample size, column totals, and row totals as the observed table. The Monte Carlo p value was then determined as the proportion of the resulting 10,000 tables that yielded a test statistic at least as large as the one observed. We chose Monte Carlo-based tests rather than traditional Fisher exact tests because the number of permutations needed to calculate the exact tests was extremely large and memory intensive. Following these univariable associations, we fit a nominal, multivariable, multi-categorical logistic regression model to determine variables independently associated with Year-4 alcohol intake. Alcohol intake was fit as the outcome, and all variables univariably associated with alcohol intake (p < 0.05) were included in the model, using a multinomial distribution and a generalized logit (glogit) link function. Because multinomial logistic regression odds ratios can be challenging to interpret, we instead presented estimated marginal percentages of participants falling within a given alcohol intake category for each level of each characteristic of interest, similar to the calculation of least squares means for continuous outcome variables. First, predicted probabilities of alcohol intake were calculated for each combination of all variables in the model. Then, for a given level of a given characteristic of interest, the mean of the predicted probabilities of alcohol intake for that level was determined by averaging all predicted probabilities for that level across all combinations of the other variables in the model. These marginal predicted probabilities were multiplied by 100 to arrive at predicted percentages.
Change in alcohol intake from baseline to Year 4 was assessed using Bowker’s test of symmetry. Following this, we examined univariable associations of changes in alcohol intake from baseline to Year 4 with characteristics of interest, again using Monte Carlo-based Fisher exact tests. Primary analyses were subset to participants with baseline alcohol intake of 1–4 or 5–14 drinks per week, since participants at the lower extreme of alcohol intake (< 1 drink per week) could not decrease alcohol consumption at Year 4 and those at the higher extreme (15 + drinks per week) could not increase alcohol consumption at Year 4. Participants were grouped into three change categories based on change in alcohol intake from baseline: decreased alcohol intake at Year 4, no change in alcohol intake at Year 4, and increase in alcohol intake at Year 4. Multivariable analyses again included all variables found to be univariably significant (p < 0.05), and resulting marginal percentages calculated, similar to the Year-4 cross-sectional analyses.
All statistical tests were two-sided, and all analyses were carried out using the SAS system (SAS Institute, Inc., Cary, NC).

Results
A total of 2230 patients with stage 0–3 BC returned a baseline questionnaire. Of these, we excluded 1502 who either did not return the Year-4 questionnaire or provided insufficient alcohol intake data at baseline or at Year 4, 2 whose BC recurred prior to the Year-4 questionnaire, and 7 who returned their baseline survey more than one year after their original diagnosis, resulting in a final cohort of 719 patients. Associations of participant characteristics with return of the Cohort 4 questionnaire, among all 2230 patients, are provided in Supplementary Table 1. Patients who were white, premenopausal, with Stage 0–2 disease, who did not undergo chemotherapy or radiotherapy, and who had higher PROMIS-10 mental and physical health scores were more likely to return the Year-4 questionnaire than their demographic and clinical counterparts.
Cohort characteristics are provided in Table 1. The mean age at BC diagnosis was 58.6 years (SD = 12.0). Ninety-nine percent of participants were female, and 97% were of White race. Most participants were postmenopausal (74%). Approximately, two-thirds of participants were never smokers (67%). More than half held a bachelor’s or graduate-level degree. One quarter of participants reported financial difficulty and 27% reported not engaging in at least mild exercise per week. Nearly half of participants had stage 1 BC (49%), and the majority of tumors were ER -positive (85%), PR -positive (77%), and HER2 -negative (88%). Slightly more than half of the participants (51%) underwent lumpectomies, 66% received endocrine therapy, 27% received chemotherapy, and 52% received radiotherapy.

Cross-sectional associations of alcohol intake at Year 4 with participant characteristics
Of the 719 participants, 217 (30%) reported little or no alcohol intake, 351 (49%) consumed 1–4 drinks per week, 134 (19%) consumed 5–14 drinks per week, and 17 (2%) reported consuming 15 or more drinks per week. Univariable associations of alcohol intake at year 4 with participants’ characteristics are provided in Table 1. Lower alcohol consumption was associated with older age at diagnosis, such that 38.5% of participants aged 70 and older reported no alcohol intake compared to 6.4% of participants < 40 (p < 0.001) (Table 1, Fig. 1). Thirteen percent of current smokers reported consuming 15 or more alcoholic drinks per week compared to 3.7% for former smokers and 1.5% for never smokers (p < 0.001). Alcohol intake was lower in participants reporting zero minutes of moderate exercise per week (p = 0.001) and in participants reporting zero minutes of mild exercise per week (p = 0.005) compared to those reporting at least some exercise. Similarly, participants with lower PROMIS global physical health scores were more likely to abstain from alcohol compared to those with higher scores: 42% with scores < 46 abstained compared to 29% with scores of 56 or greater (p = 0.028). No significant associations were observed between alcohol intake and tumor-related or treatment characteristics (p > 0.05 for each).
Age at diagnosis remained a significant correlate of alcohol consumption in multivariable analyses, in that 6.6% of participants less than age 40 reported no alcohol intake compared to 38.1% of participants 70 and older (p = 0.022, Fig. 1a, Supplementary Table 2). Smoking also remained significantly associated with alcohol consumption, such that 22% of current smokers abstained and 17.5% consumed 15 or more drinks per week compared to 33.4% and 1.6% for never smokers (p < 0.001, Fig. 1b, Supplementary Table 2). Exercise and physical health scores were no longer significantly associated with alcohol intake in multivariable analyses.

Trend in alcohol intake from baseline to Year 4
The proportion of participants falling into each alcohol intake category at Year 4 was not significantly different from that at baseline (Bowker test for homogeneity p = 0.13), 27.7% reported no alcoholic drinks per week at baseline compared to 30.2% at Year 4, 48.3% reported 1–4 weekly drinks at baseline compared to 48.8% at Year 4, 21.0% reported 5–14 drinks per week at baseline compared to 18.6% at Year 4, and 3.1% reported 15 + drinks per week at baseline compared to 2.4% at Year 4 (Table 2). Of the 719 participants, 73 (10.2%) increased alcohol intake from baseline to Year 4, 112 (15.6%) decreased intake, and 534 (74.3%) did not change their intake.

Associations of change in alcohol intake from baseline to Year 4 with participant characteristics
Univariable associations of change in alcohol use from baseline to Year 4 with demographic and clinical characteristics, among the 498 participants not at the lowest or highest levels of baseline alcohol use, are presented in Table 3. Increases in alcohol consumption were more common in participants reporting 60–119 (13.0%) and 120 + minutes of strenuous exercise per week (15.9%) compared to those reporting 0 (7.5%) or 1–59 min per week (6.9%, p = 0.011). Conversely, decreases in alcohol consumption were more common in participants reporting 0 (21.0%) or 1–59 min of strenuous exercise per week (34.5%) compared to those reporting 120 + minutes per week (7.2%, p = 0.011). Similarly, participants reporting 240 or more weekly minutes of mild exercise per week were more likely to increase their alcohol consumption (18.4%) compared to those with 0 (9.7%), 1–119 (7.0%) or 120–239 min per week (4.6%, p = 0.012). A lower proportion of participants who received endocrine therapy decreased their alcohol consumption from baseline to Year 4 (16.8%) than those who did not receive therapy (25.6%, p = 0.047). No other demographic or clinical characteristics were significantly associated with changes in alcohol consumption (p > 0.05 for each). In multivariable analyses, the only attribute associated with change in alcohol consumption was weekly minutes of mild exercise per week (p = 0.026, Fig. 2, Supplementary Table 3). Results were similar to those seen in univariable analyses, in that participants reporting lower levels of exercise were more likely to decrease their alcohol consumption compared to those reporting higher levels, and those reporting 240 or more minutes of mild exercise per week were more likely to increase consumption compared to those reporting lower exercise levels.
Univariable sensitivity analyses re-examining associations of change in alcohol use after including all 719 participants (without censoring those at the extremes of alcohol use) are provided in Supplementary Table 4. Associations of change in alcohol use with strenuous exercise, mild exercise, and receipt of endocrine therapy were similar to those found in primary analyses. Increase in alcohol intake was also univariably associated with mastectomy (vs. lumpectomy only) and with PR-positive and HER2-positive disease status. None of these variables retained significance in the subsequent multivariable model.

Discussion
In this prospective cohort study of BC survivors, we examined clinical, demographic, lifestyle, and socioeconomic factors associated with alcohol consumption 4 years after diagnosis, as well as changes in alcohol use from the time of diagnosis to late survivorship. Our group has previously found that in early survivorship (Year 1), participants overall reduced alcohol consumption compared to baseline, with the proportion of participants consuming < 1 drink/week increasing from 28.9% at baseline to 60.2% at Year 1, and those drinking ≥ 1 drink/week declining from 71.2% at baseline to 39.8% at Year 1 [19]. Our new findings extend our prior research by evaluating behavior at a later point in the survivorship trajectory. This later timepoint captures a distinct and underexamined phase in survivorship, during which behavior patterns may stabilize or shift in response to long-term psychosocial adjustment, evolving health status, and changes in care intensity.
Consistent with prior studies in cancer populations, younger age at diagnosis was strongly associated with higher levels of alcohol intake in both univariable and multivariable cross-sectional analyses of alcohol use at Year 4. Studies have shown that cancer survivors < 65 years of age are more likely to engage in current alcohol consumption, to exceed moderate drinking, and to engage in binge drinking compared to those aged 65 years and older [21, 22]. These findings highlight the need for targeted interventions to address alcohol consumption and support sustainable behavioral change among younger cancer survivors, given the potential adverse effects on treatment outcomes and overall health.
Current smoking status remains strongly and consistently associated with higher alcohol intake at Year 4 and across all timepoints including in our prior studies at baseline and Year 1. This is consistent with well-documented clustering of health risk behaviors. Continued smoking through several timepoints in survivorship may be influenced by comorbid psychological distress, anxiety, and depression as well as insufficient cessation support [23, 24]. This dual-risk profile of combined alcohol and tobacco use has particular relevance for survivorship care, as combined use is associated with worse cancer outcomes such as recurrence and development of subsequent primary cancers as well as overall mortality [25, 26]. Identifying individuals with multiple behavioral risk factors highlights the need for increased psychologic support over time for affected patients and can help integrate behavioral counseling approaches more effectively.
In a multivariable model, survivors reporting higher levels of mild or strenuous exercise were found to be more likely to increase alcohol intake, while those with lower activity levels were more likely to reduce consumption. This finding mirrors the literature suggesting that some individuals may compensate for alcohol consumption through healthy behaviors such as exercise or may socialize more frequently in physically active settings that promote alcohol use [27–29].
Interestingly, although univariate analyses identified an association between physical health (via PROMIS-10 global physical score) and alcohol use, with less alcohol intake seen among participants with little or no physical activity, these associations did not persist in multivariate models. This may reflect the interplay between lifestyle behaviors and other health comorbidities that confound these associations. Those with poorer physical function may be less able to engage in physical activity and more likely to abstain from alcohol due to medication contraindications or long-term comorbid conditions.
Treatment-related variables were not significantly associated with alcohol intake in cross-sectional models. This differs from analyses at Year 1, which revealed an association between receipt of chemotherapy and a reduction in alcohol intake (possibly due to chemotherapy-related taste-disturbances and nausea). This may reflect a decoupling of medical and behavioral trajectories as patients transition from active treatment into survivorship.
Our study also found that most BC survivors in our cohort (74.3%) reported a similar amount of alcohol consumption at baseline and at 4 years following diagnosis. Only 15.6% reported a decrease and 10.2% reported an increase in intake. This is a significant difference from the declines seen in alcohol use in this cohort one year after diagnosis. This suggests that the greatest behavior shifts may occur early post-diagnosis, with subsequent trajectories either stabilizing or partially reverting over time, and that initial post-diagnosis changes in behavior may not be sustained over time. This finding aligns with existing literature on “teachable moments” in cancer care, highlighting that early survivorship may be a limited window in which survivors are most receptive to lifestyle change [30, 31]. The adoption of healthier behaviors during and soon after cancer treatment may not be sustained in the long term as initial motivations wane, routine medical oversight lessens, and long-term coping patterns may change [32]. This underscores the importance of addressing alcohol use as a dynamic, long-term behavior, and the need for repeated counseling throughout survivorship with regular reassessment and coordinated care between oncology and primary care providers to ensure sustained lifestyle improvements and better long-term health outcomes.
Strengths of this study include its longitudinal design, a large sample size, comprehensive collection of clinical, demographic, and lifestyle data, and its focus on a relatively unexplored phase of survivorship. To our knowledge, this is among the first studies to explore predictors of alcohol intake in BC survivors several years after diagnosis, providing insight into long-term behavioral patterns and their determinants. While our findings have important implications, they must be interpreted considering potential limitations. Supplemental analyses indicate that patients who were White, who were premenopausal, who had lower-stage disease, who did not receive chemotherapy or radiation, and who reported higher physical and mental health scores were more likely to complete the Year-4 survey. These differences suggest that our sample may underrepresent survivors with more intensive treatment histories or poorer health status. The study is limited by the lack of detailed information on the type of alcohol consumed (such as wine, beer, or spirits) and the context of consumption (e.g., with meals or binge drinking). Additionally, alcohol intake was assessed as average weekly consumption over the preceding 10 years at both baseline and Year 4 timepoints, such that these periods overlapped, limiting our ability to detect differences between pre-cancer intake and intake during survivorship. This means that we may be underestimating behavioral shifts. Furthermore, self-report of alcohol intake may have led to recall and social desirability bias, producing underreporting of alcohol intake. Our sample was also demographically homogeneous (largely White and well-educated), limiting generalizability to more diverse populations.
In conclusion, many BC survivors drink alcohol 4 years after a BC diagnosis (a greater proportion than we found previously at one year after a BC diagnosis) [19]. These findings highlight the importance of continued, nuanced health behavior counseling in both early and late survivorship care, particularly targeting younger survivors, smokers, and those with more active lifestyles (perhaps because they may not perceive their alcohol intake as a health risk). Further research is warranted to explore the motivations underlying changes in alcohol use during cancer survivorship and to develop tailored interventions to support long-term health in this population.

Supplementary Information
Below is the link to the electronic supplementary material.