Introduction

Breast cancer (BC) is the first cause of cancer death among women. Indeed, the prevalence of BC is the leading cause among all types of cancers [1]. Although genetic or environmental factors can be responsible for BC development, available evidence suggests that diet may be one of the most important determinants of BC occurrence [2]. In fact, several studies have highlighted the importance of lifestyles in the development of BC, which are also directly related to pathophysiological mechanisms present in obesity, metabolic syndrome and/or cardiovascular disease [3, 4].

Hitherto, coffee consumption has been suggested to decrease risk of Parkinson’s disease, type 2 diabetes, cardiovascular diseases and breast, colorectal, endometrial, prostate cancers and even all-cause mortality [5,6,7,8]. Coffee consumption has also been linked to a lower risk of premenopausal BC [9,10,11] and postmenopausal BC in different cohort studies, among them, 21 cohort studies were included in a recent meta-analysis [9, 12,13,14]. However, none of them adjusted for an overall dietary pattern although coffee consumption may be associated with a holistic dietary pattern [10].

Some effects of coffee consumption on chronic disease have been traditionally attributed to caffeine, an alkaloid widely present in coffee [15]. However, the association between caffeine intake and BC risk in observational studies is still inconsistent; while one study found no significant association [16], others did [9, 12]. On the other hand, coffee also contains other minor components such as diterpenes, polyphenols like chlorogenic acids, melanoidins, N-methylpyridinium and acrylamide, whose concentrations depend on the handling and the variety of coffee studied [17, 18]. The potential biological actions of these minor components—which may interact in synergy—on the prevention of BC ranged from regulation of DNA repair, proapoptotic, antiproliferative, antiangiogenic and anti-metastatic effects to antioxidant properties as regulation of Nrf2 and Kelch-like ECH-associated protein 1 [5, 19]. In fact, coffee is the main source of polyphenol intake in several populations in which the aforementioned studies on the association between coffee consumption and BC were conducted [20]. However, no previous study on this association has been conducted in countries such as Spain in which the main source of polyphenol intake is fruits and vegetables [21, 22] and therefore, the association between coffee consumption and BC risk may be different.

We investigated whether coffee consumption is associated with BC overall and we stratified the results by menopausal status in a prospective cohort of Spanish university graduates, after adjusting for the overall dietary pattern.

Materials and methods

Study sample

The “Seguimiento Universidad de Navarra” (SUN) project is a prospective, multipurpose cohort of Spanish university graduates. The study methods have been described in more detail elsewhere [23]. Briefly, the SUN project is a dynamic cohort where the recruitment is permanently open assessing the relationship between diet and chronic diseases. It was developed inspired by the models of the Nurses’ Health Study and the Health Professionals Follow-Up Study. Recruitment started in December 1999.

After the initial questionnaire, follow-up questionnaires are mailed every 2 years to participants to update information on diet and lifestyle and collect information on health outcomes which might have happened in the previous 2 years. For participants lost to follow-up, the National Death Index is consulted periodically to assess their vital status and cause of death. Participants are middle-aged university graduates from different Spanish regions.

By 2018, 22,791 participants were recruited. To allow the minimal follow-up of 2 years, we included only those participants who were recruited before October 2015 (2.75 years before the database closing date). Out of 13,770 eligible women, we excluded 1295 participants with no follow-up information (retention 90.6%); 104 participants with self-reported previous history of BC at baseline, 1353 participants with total energy intake out of predefined limits (< 500 or > 3500 kcal/day) [24], or reported menopause before the age of 35 years (n = 206). The final sample consisted of 10,812 participants who answered at least one follow-up questionnaire.

This study was conducted according to the guidelines laid down in the Declaration of Helsinki, and all procedures involving participants were approved by the Institutional Review Board of the University of Navarra. Voluntarily given informed consent through freely fulfillment of the baseline questionnaire were gathered from all participants according to the methods approved by our Institutional Review Board.

Assessment of coffee consumption

The baseline questionnaire included a previously validated 136-item food-frequency questionnaire (FFQ) [25, 26]. The reproducibility of the FFQ has been specifically addressed in this cohort [27]. The serving size for coffee was 50 cc or a cup size. Information about the consumption of regular and decaffeinated coffee was gathered separately. The FFQ assessed habitual food consumption over the previous 12 months and included nine categories of response for the frequency of consumption, ranging from ‘never/seldom’ to ‘more than six times per day’. Then, participants were grouped in two categories according to their level of coffee consumption (≤ 1 cup/day and > 1 cup/day). Consumption of regular and decaffeinated coffee was added to quantify total coffee consumption.

Ascertainment of incident breast cancer cases

For the present analysis, incident BC was the primary endpoint. Participants were asked whether they had received a medical diagnosis of BC at baseline and during follow-up. Incident BC was defined as women who developed BC during the follow-up and were free of the disease at the beginning of the study. They were also inquired about the date of diagnosis. Self-reported BC cases were considered as probable cases. These participants were asked for a copy of their medical records. Then, a trained oncologist confirmed the cases. Moreover, deaths were reported to the research team by subject’s next of kin, work associates and postal authorities. If lost during the follow-up or with unidentified causes of death, the National Death Index was consulted periodically to identify deceased cohort members and confirm the cause of death. Both cases adjudicated by a trained oncologist and deaths due to BC were considered as confirmed BC cases (n = 101) (false positives, n = 7). These trained oncologists were blinded with respect to dietary exposures.

Assessment of covariates

Mediterranean diet adherence was assessed using the score proposed by Trichopoulou et al. with some modifications [28], which consisted of using an 8-point Mediterranean-diet scale ranging from 0 to 8 (because we removed the moderate alcohol component due to the postulated detrimental effect of alcohol on BC). Participants received one point if they were above the median for the consumption of each of six typical elements of this dietary pattern (vegetables, legumes, fruits and nuts, cereal, fish and the monounsaturated/saturated ratio intake), and another point was added if they were below the median for the consumption of two items that were not included in the traditional Mediterranean diet (red meat and meat products and dairy products). Coffee consumption was not considered as part of this score.

The baseline questionnaire also included information about participants’ socio-demographic characteristics, medical history (prevalent chronic diseases), health-related habits, lifestyles, anthropometric data, physical activity and family history of BC [29, 30]. We included as ultraprocessed foods the total amount of the food items meeting the ultraprocessed food category of the NOVA system [31] petit suisse, breakfast cereals, milkshakes, custard, pudding, ice-cream, potato chips, cookies, muffins, donuts, croissant, cakes, churros, chocolates, nougat, marzipan, ready-to-consume pies, pizza, instant soups and creams, margarine, mayonnaise, sugar sweetened beverages, diet soft drinks, bottled fruit juice, processed meat (cooked ham, spicy pork sausage, salami, mortadella, foie-gras, black pudding, bacon, hamburger, sausage), and alcoholic drinks produced by fermentation followed by distillation such as whisky, gin, rum or vodka.

Statistical analysis

Baseline quantitative traits of participants were described as the means and standard deviations according to categories of coffee consumption and baseline qualitative traits as the percentages across the same categories.

We assessed the correlation between coffee consumption and other beverage consumption with Pearson’s correlation coefficients.

To assess the relation between baseline total coffee consumption (categorized as ≤ 1 cup/day and > 1 cup/day) and the subsequent risk of BC, we used Cox regression models with age as underlying time variable, stratified the Cox regression models for age (decades) and recruitment period (three periods) and included confirmed incident cases as the outcome. Baseline regular coffee was separately assessed. Regular coffee consumption was categorized as ≤ 1 cup/day and > 1 cup/day. Additionally, baseline regular and total coffee consumption were separately assessed as continuous variables, to describe the association according to the change in one cup of regular or total coffee consumption. We defined follow-up for each participant from the date of completion of the baseline questionnaire to the date of BC diagnosis for cases and to the last returned questionnaire for non-cases. Probable incident cases were considered as non-cases and censored at the date of self-reported diagnosis. Hazard ratios (HR) and their 95% CI were calculated considering the lowest category of coffee consumption as the reference category.

For all analyses, we fitted several multivariable models with successive degrees of adjustment for potential confounders. Model 1 was adjusted for height (continuous), family history of BC (yes/no), smoking status (never smoker, former smoker, current smoker), smoking package-years (continuous), physical activity (MET-h/week, continuous), alcohol intake (g/day, continuous), BMI (continuous), age of menarche (< 10 years, 10–11 years, 12–13 years, 14–16 years, > 16 years), menopause (no menopause, < 50 years, ≥ 50 years), number of pregnancies of more than 6 months (continuous), pregnancy before the age of 30 years (yes/no), months of breastfeeding (continuous), use of hormone replacement therapy (yes/no) and its duration (continuous) and years at university (continuous).

Model 2 included additional adjustment for prevalent diabetes (yes/no), BMI (quadratic), total energy intake (kcal/day, continuous; linear and quadratic terms), ultra-processed food consumption (servings/day), sugar-sweetened beverages consumption (yes/no) and adherence to the Mediterranean diet (continuous).

Dietary data were updated after 10 years of follow-up to lessen the potential effect of diet variation during follow-up, and time-varying Cox proportional hazard models were fitted with repeated measures of the coffee intake.

Subsequently, we ran stratified analysis among pre- or postmenopausal women and their risk of BC. Information on age at menopause was collected in the baseline questionnaire and after 16 years of follow-up. If information regarding the age at menopause was lacking, we imputed postmenopausal status according to the 75th percentile of age at menopause in the study population (52 years of age) (n = 3132) [32]. When assessing premenopausal BC as the outcome, we excluded those women who reported having had menopause before study inception and censored follow-up time at the age of 52 years or at the self-reported age of menopause, whichever happened first. For postmenopausal BC, we considered women at risk only those after having turned 52-year-old or after the self-reported age of menopause, whichever happened last. In the analysis with postmenopausal BC as outcome, we additionally adjusted for age at menopause (< 50 years, ≥ 50 years) and time since recruitment until the beginning of the time at risk. The latter was 0 for women who were already postmenopausal at study inception. For women who were initially premenopausal but turned postmenopausal during follow-up, time since recruitment was calculated as the difference between the initial date of being at risk of postmenopausal BC and the date of completion of the baseline questionnaire.

To assess effect modification, differences in the association of the exposure with breast cancer by menopausal status were evaluated with interaction and with the likelihood ratio test. The p value for interaction by menopausal status was calculated by comparing models with or without a multiplicative interaction term between menopausal status and coffee consumption.

Finally, based on the question “Do you add sugar to some beverages?”, we separated coffee consumption of those participants who added sugar from those who did not and assessed the interaction between adding sugar to coffee or not in their relationship to BC with the likelihood ratio test.

As sensitivity analyses, we repeated our analyses with probable BC cases and explored the robustness of our results using other cut-offs criteria for energy intake, and using total coffee intake as the main exposure (Fig. 1). Particularly, we used the basal metabolic rate (BMR) and excluded people under BMR*1.2 and/or over BMR*1.9 using the Miffil-St Jeor equation [33] or under or above the percentiles 5 and 95, respectively, of the ratio between total energy intake and BMR. We also performed further analyses using different exposure cut-off points (< 1 cup/day, 1 cup/day, > 1 cup/day).

Fig. 1
figure 1

Sensitivity analyses. Hazard ratios and 95% CI of incident BC for drinking > 1 coffee/day vs. ≤ 1 coffee/day according to menopausal status and different energy cutoff points

Analyses were carried out using Stata version 12.0 (Stata Corporation). All p values were two-tailed and a p value < 0.05 was deemed as statistically significant.

Results

The main baseline characteristics of the 10,812 women included in our analyses according to categories of baseline total coffee consumption are shown in Table 1. The median age of participants was 32 years and the mean BMI was 22.2 kg/m2 (SD 3.1). Participants in the highest category of total coffee consumption were older and less physically active, had a higher BMI, higher total energy intake and intake of alcohol, meat, fruits, vegetables, olive oil and dairy products, higher adherence to the Mediterranean diet and lower cereal consumption compared to women with 1 or less cup/day of coffee consumption.

Table 1 Baseline characteristics of 10,812 women in the SUN cohort according to categories of total coffee consumption

Coffee consumption did not show strong inverse correlation coefficients with any other beverage (Supplementary Table 1).

Incidence of breast cancer

We identified 189 probable incident cases of BC. Out of these, we confirmed 101 new-onset cases of BC among 115,802 person-years of follow-up (median follow-up 11.8 years).

We observed no significant association between total coffee consumption and overall BC risk when we considered either confirmed cases (adjusted HR 1.11; 95% CI 0.74, 1.66) (Table 2) or probable cases (adjusted HR 1.10; 95% CI 0.82, 1.48) (Supplementary Table 2). Results did not change substantially when we set other cut-off points for total energy intake (Fig. 1). Adding sugar to beverages did not modify the described association.

Table 2 Hazard ratio (HR) and 95% confidence intervals (CI) of confirmed breast cancer cases according to the categories of baseline coffee consumption among 10,812 women of the SUN Project

We did not observe any association with overall BC risk when we considered regular coffee consumption (adjusted HR 1.20; 95% CI 0.80, 1.79) (Supplementary Table 3), or when we addressed this association including also probable cases (Supplementary Table 2).

Incidence of premenopausal and postmenopausal breast cancer

When we divided participants according to menopausal status and considered coffee consumption of less or equal than 1 cup/day as the reference category, consumption of more than 1 cup/day of coffee among premenopausal women was associated with a non-significant increased risk of BC (adjusted HR 1.69; 95% CI 0.96, 2.96) (Table 3). The associations remained non-significant when we considered only regular coffee consumption (Supplementary Table 4) or when we included also probable cases of premenopausal BC (Supplementary Table 5). Results barely changed in sensitivity analyses concerning energy intake limits (Fig. 1) or other exposure cut-off points as shown in Supplementary Table 6.

Table 3 Hazard ratio (HR) and 95% confidence intervals (CI) of confirmed breast cancer cases for each category of coffee consumption among premenopausal and postmenopausal women of the SUN project

On the other hand, postmenopausal women who consumed more than 1 cup/day of coffee showed a significantly lower risk of BC (adjusted HR 0.44; 95% CI 0.21, 0.92) in the fully adjusted model compared to postmenopausal women in the lowest consumption category (Table 3). Results were consistent when we considered only confirmed cases or all probable cases (Supplementary Table 5) as well as when we set other exclusion criteria (Fig. 1) or other exposure cut-off points in the sensitivity analyses (Supplementary Table 6).

No significant association was found between regular coffee consumption (adjusted HR 1.35; 95% CI 0.92, 1.97) and postmenopausal BC risk when we considered all cases, including also probable cases (Supplementary Table 5). When we restricted the analysis to confirmed cases, we could only estimate the association between regular coffee consumption and postmenopausal BC risk which was not statistically significant (Supplementary Table 4).

When the analyses were updated with repeated measures, the results remained similar and were statistically significant for postmenopausal women (adjusted HR 0.43; 95% CI 0.21, 0.90).

The interaction between coffee consumption and menopausal status in their association with BC was statistically significant (p value = 0.008). There was a statistically significant inverse association between coffee consumption and BC among postmenopausal women (HR:0.44; 95% CI 0.21, 0.92) whereas the association was not statistically significant for premenopausal women (HR: 1.69; 95% CI 0.96, 2.96).

Discussion

In this prospective study of Mediterranean university graduates, we observed no association between coffee consumption and overall breast cancer risk. In ancillary analyses, according to menopausal status, we found a lower risk of BC among postmenopausal women who consumed more than one cup of coffee per day compared to those who consumed one cup or less per day, even after adjustment for the overall dietary pattern. Moreover, a statistically significant interaction between menopausal status and BC incidence associated to coffee consumption was found. However, no significant association between coffee consumption and BC was observed among premenopausal women. What is more, the point estimate for the association among premenopausal women was towards the opposite direction and our results were compatible with a potential detrimental association among premenopausal women.

This inverse association between coffee consumption (> 1 cup/day versus lower consumption) and postmenopausal BC incidence is consistent with previous literature. Lafranconi et al. pooled a total of 21 prospective studies and observed that women who consumed at least two cups of coffee per day showed a lower risk of BC compared to non-consumers and the association was stronger for postmenopausal women [12]. Li et al. conducted another meta-analysis in which they also distinguished between estrogen receptor-positive (ER+) and estrogen receptor-negative (ER−) BC. They included 23 studies and found a borderline significant influence of highest coffee consumption, which corresponds to 2 cups/day increment in coffee intake, (RR 0.98; 95% CI 0.97–1.00) on the risk of overall BC but an inverse association between coffee consumption and ER-BC [34].

Regarding the different types of coffee, regular coffee consumption was not significantly associated with overall, premenopausal or postmenopausal BC risk. Indeed, no significant association between either caffeinated or decaffeinated coffee and BC risk was observed in the NIH-AARP Diet and Health Study cohort [35]. However, regular coffee intake was associated with lower risk of postmenopausal BC (HR 0.90, 95% CI 0.82–0.98) in the EPIC study [36]. Nevertheless, according to current data, there is no sufficient evidence to support the idea that regular coffee consumption is more strongly associated with BC occurrence than decaffeinated coffee. Thus, it seems that beneficial properties of coffee against BC risk cannot be attributed only to caffeine.

Coffee is composed by several and distinct constituents besides caffeine: diterpenes (coffee lipids), phenolic acids (polyphenols), melanoidins, N-methylpyridinium and acrylamide. Concentrations of these single components may vary due to coffee variety, handling, effect of time, roasting temperature and preparation method [37,38,39,40] and, consequently, the effects of coffee consumption may also be different according to the type of coffee. The type of coffee might be important due to the quantity of polyphenols and diterpenes that could vary depending on it [37, 41, 42]. But, even in minor quantity, these compounds are found in all types of coffee. So, chlorogenic acids may regulate oxidative and inflammatory stress conditions [43], although these effects have been described to decrease with increasing roasting [38]. Regarding the potential health effects of these minor components, several of these coffee components, such as chlorogenic acid, trigonelline, kahweol, and cafestol, have been found to exert antioxidant and anti-inflammatory properties [44,45,46] which may prevent tumor appearance. Even more, some of these compounds (including polyphenols) have been described to possess antitumor and antimetastatic effects by inducing cell cycle arrest or promoting apoptosis of cancer cells (i.e., via PI3K-Akt) [34, 47, 48].

Our study has various potential limitations. First of all, our statistical power could be suboptimal, because of the reduced number of incident cases of BC. Our cohort is largely constituted by young women, which diminishes the number of incident BC cases. In fact, we were not able to estimate the association between decaffeinated coffee consumption and the risk of postmenopausal BC due to the low number of cases, nor were we able to address the association between a greater number of categories of coffee consumption and BC. Second, self-reported information used on coffee consumption may imply some degree of misclassification which in turn, may bias our results towards the null. However, dietary information was collected with a previously validated semi-quantitative food-frequency questionnaire [26, 27]. Third, coffee consumption was assessed only at baseline in the main analyses, but this may not represent a strong limitation given that coffee consumption tends to remain stable over the years and we also repeated our analyses with updated information on coffee consumption after 10 years of follow-up [49, 50]. Fourth, the lack of information regarding the method of preparation, type and variety of coffee could be a strong limitation; however, previous studies reported that non-filtered coffee is the main type of coffee consumed in Spain. This latter variety includes coffee prepared with the use of pressure (espresso coffee), a percolator (a type of pot that brews coffee by passing boiling water over the grounds), and instant coffee [49]. Fifth, information on BC incidence was self-reported. We acknowledge that we might have missed some BC cases. However, to avoid false positives, BC cases were confirmed by a trained oncologist. Sixth, we did not have a representative sample of the general population. However, from the point of view of biological plausibility, the association between coffee consumption and BC risk does not have to be different for highly educated women than for women with a lower educational level. Moreover, most cohorts and other analytical studies are usually non-representative, and this represents no problem given that the generalization should be based on biological plausibility. Seventh, postmenopausal status was imputed and derived by censoring which might result in classification errors. Eighth, since tea was not widely consumed in Spain by the time the FFQ was developed, tea was not selected as a relevant item to be included in the FFQ and we had no information on tea consumption. Ninth, we aimed to assess if adding sugar to coffee significantly modified the results. Nevertheless, this question was not limited to coffee consumption but included all beverages and we did not have information on artificial sweeteners. In fact, we could not distinguish if sugar—or similarly milk—were poured into coffee or into other beverages since dietary information was collected with an FFQ.

Nonetheless, our study also has some strengths, starting from its longitudinal design that reduces the possibility of reverse causation bias and the adjustment for many potential confounders which have been included in the multivariable analyses, although we cannot completely rule out the possibility of residual confounding. Biological plausibility is described with anti-inflammatory and antioxidant effects of coffee compounds, and, furthermore, coffee has been described to exert antitumoral effects in several types of cancers besides BC.

In conclusion, according to our results there are slight indications of an inverse association between coffee consumption and postmenopausal BC risk. These findings are consistent with previous studies and support the idea that coffee could be incorporated in a healthy diet.