Dose-Dependent Associations between Wine Drinking and Breast Cancer Risk-Meta-Analysis Findings

Breast cancer is an important public health issue, as it is the leading malignancy with high incidence and mortality among women globally (Arveux and Bertaut, 2013). Several risk factors, such as first-degree family history, breast cancer susceptibility gene 1 (BRCA1) and BRCA2 mutations, were identified and related to breast cancer (Espie et al., 2013). Wine, as a special type of alcohol beverage, contains more than one chemoprotective chemical, including iso-flavone phytoestrogens, flavones, and procyanidin B dimmers (Eng et al., 2003; Key et al., 2006). In 2007, the International Agency for Research on Cancer (IARC) classified alcohol as carcinogenic to several human malignancies (Seitz and Stickel, 2007). Since then, the association between alcohol and breast cancer risk attracted much attention. Several epidemiological studies have demonstrated that alcohol consumption was associated with an increased risk of breast cancer (Smith-Warner et al., 1998; Corrao et al., 1999; Ellison et al., 2001; Singletary and Gapstur, 2001; Hamajima et al., 2002; Chen et al., 2011). However, results


Introduction
Jia-Yan Chen 1& , Hong-Cheng Zhu 1& , Qing Guo 2& , Zheng Shu 3 , Xu-Hui Bao 4 , Feng Sun 5 , Qin Qin 1 , Xi Yang 1 , Chi Zhang 1 , Hong-Yan Cheng 6 , Xin-Chen Sun 1 * from studies seem controversial (Willett et al., 1997;Higgins et al., 2003;Bessaoud and Daures, 2008). Some issues about alcohol and breast cancer still remain complex and not well understood, such as the different effect of beverage choice (wine, liquor or beer). There was little evidence on whether different types of alcoholic beverage, including wine, liquor, and beer, play similar roles.
Meanwhile the dose-risk relation of wine intake with breast cancer hasn't yet been completely studied in detail. In particular, it is still not clearly established whether low dose wine consumption was associated with protective effect on breast cancer. It seems that more precise quantification and identification of a possible threshold for effect of wine are needed to be decided.
We herein performed a dose-response meta-analysis to investigate the potential association between wine and breast cancer risk.

Search strategy and selection criteria
Medline, Embase, and the Cochrane Library were searched from inception to May 8th 2015 with the following subject heading terms and/or text words: "breast cancer", "breast neoplasm" in combination with "wine", "alcohol", "drinking", "beverage". In addition, a broader search on diet and breast cancer was also conducted. Further, the reference lists of retrieved articles and relevant review articles were scanned. No language restrictions were imposed. According to "Food, nutrition, physical activity, and the prevention of cancer: a global perspective", wine was defined as alcoholic drinks produced from grapes and contain between around 9 to 15 per cent alcohol; The composition of wine depends on the grape varieties used, including red wine, white wine, sparking wine, et al (Wiseman, 2008). Studies were included if they (i) had a case-control or cohort design; (ii) evaluated the association between wine drinking and breast cancer risk; (iii) presented odds ratio (OR), relative risk (RR) estimates with 95% confidence interval (CI). If publications were duplicated or articles from the same study population, the publication with a larger scale was included. Nonpeer-reviewed articles, ecologic assessments, correlation studies, experimental animal studies and mechanistic studies were excluded. All the process was conducted by two independent investigators (Jiayan Chen and Hongcheng Zhu).

Data extraction and quality assessment
Two independent investigators (Jiayan Chen and Hongcheng Zhu) extracted the following data from each study that met the criteria for inclusion: first author, year of publication, geographic regions, journal, number of cases, cohort size, cohort name and duration of follow-up (cohort studies), number and type of control subjects (case-control studies), type of cancer, consumption categories, adjusted ORs, or RRs with 95%CI, and adjusted variables. When several risk estimates were presented for pre-and postmenopause, year group, and et al. the detailed information was also extracted.
A 9-star system on the basis of the Newcastle-Ottawa Scale was used to assess the study quality from 3 broad perspectives . Considering that there is possibly a direct or indirect caloric intake with breast cancer risk, an energyadjusted residual or nutria-density model was added as an item for the scoring system (Willett et al., 1997). Hence, the full score was 10 stars, and a study with ≥7 awarded stars was defined the high-quality study (Willett et al., 1997).

Data synthesis and statistical analyses
RRs with 95% CIs were calculated using randomeffects model. ORs were considered to be equivalent to RRs since breast cancer is a rare outcome. If association estimations were provided separately from subtypes or age group of cancer, combined RRs with 95% CIs were used in the overall analysis.
Statistical analyses based on comparison of the highest intake category with the lowest intake category (which included people do not drink) were conducted. Subgroup analyses were conducted by study quality, study design (cohort studies and case-control studies), control source (population-based and hospital-based), menopause, geographic region (Europe and North America), country (Italy, France, USA, and Canada) and study adjustments (family history, body mass index, total energy, other alcohol/beverage, smoking, menopause, hormone therapy, pregnancy, and education).
In addition, categorical dose-response regression analysis was utilized. The fixed-effects linear model was first used and non-linearity test was checked. Otherwise, Flexible nonlinear meta-regression models were used. The amount of wine consumption was converted into grams of ethanol per day using the following equivalencies: 1 drink=12.5g, if not otherwise specified in the original report; 1 ounce=28.35g. Midpoint of the range of categories reported in the original reports was assigned as levels of wine consumption, and for open-ended upper categories, as 1.2 times its lower bound. Wines are estimated as 12v/v of ethanol approximately according to the majority products in the market.
Heterogeneity among studies were examined using the chi-square test, defining a significant heterogeneity as a P value <0.10 and quantified the inconsistency using the I-squared statistic (Higgins et al., 2003). Publication bias was evaluated by generating funnel plots and the Egger's test (Egger et al., 1997).

Subgroup analysis
The subgroup analysis on geographic area showed an RR of 1.66 ( . When data were adjusted by some confounding factors (family history, body mass index, total energy, other alcohol beverage smoking, menopause, hormone therapy, pregnancy, education, physical activity), the association was still statistically significant (Table 5).

Dose-response analysis
Furthermore, dose-response meta-analyses were conducted. Most of the slope of each study was greater than 0, indicating that more wine consumption might  Cohort, 8y (1976Cohort, 8y ( -1984 breast cancer None 1.0 (Referent) Five-year age categories, dummy variables for beer, wine, and liquor with "no alcohol" as the common reference group a RR = relative risk (rate ratio or hazard ratio); CI = confidence interval; BMI = body mass index; TNBC = triple-negative breast cancer; ER = estrogen receptor; lead to higher risk of breast cancer ( Table 6). The fixedeffects model was first used. The heterogeneity between studies was detected. Therefore, a random-effect model was implemented next. Table 7 showed the combined RR was 1.0059 (95%CI=0.9670-1.0464, p=0.6156) for overall meta-analysis, indicating a 0.59% increase in the risk of breast cancer for each increment of 1g per day ethanol from wine under random effect model. The test of non-linearity was significant (χ2=1763.9 P<0.0001), thus a non-linear dose-response model was performed. Figure 3 and 4 illustrated RR variation of breast cancer according to curvilinear thresholds of regular ethanol/ wine consumption. Wine was associated with breast cancer in a dose-dependent manner. The risk decreased when women who consumed below 10g (ethanol) / 80g (wine) [<1 standard drink] per day. The risk declined to the bottom at the threshold of 5g/d of ethanol and 40g/d of wine, respectively. Figure 5 shows the contour-enhanced funnel plot of studies on the association between wine and breast cancer

Discussion
To our knowledge, our meta-analysis, for the first time, evaluated the dose-response relationship between exposure to wine and risk of breast cancer. Our comprehensive metaanalysis indicated that wine consumption may increase the risk of breast cancer. However, when evaluating women drinking wine in different dosages, we found that a low dose may have some protective effect rather than an increased risk in heavy drinkers.
Consistent with many other studies, wine drinking is associated with increased risk of breast cancer risk in the highest versus lowest model. Wine, as a specific alcoholic drink, contains ethanol, which contribute to cancer risk in many published articles. The effects of ethanol may be mediated through the production of prostaglandins, lipid per-oxdation, and the generation of free radical oxygen species. Ethanol also acts as a solvent, enhancing penetration of carcinogens into cells (Wiseman, 2008). Interestingly, a recent study suggested that low dose of wine intake can decreased the risk of breast cancer (Bessaoud and Daures, 2008). In this study, the risk associated with women who consumed wine at low dose also showed decreased tendency at a non-linear dose-response model. The protective effect of low dose wine consumption on breast cancer is plausible for several reasons. First, wine contains high levels of anticancer compounds, such as polyphenols and resveratrol. A preclinical study tested the anti-proliferative activity of these compounds on the proliferation of different breast cancer cell lines, showing that low concentrations (nanomolar or even the picomolar range) of these active <50 g/wk 0.8 (0.6-1.0) 50-149 g/wk 0.9 (0.6-1.2) ≥ 50 g/wk 1.2 (0.8-1.9) a OR = odds ratio; CI = confidence interval; BMI = body mass index; BRAC = ; T = tertile; BRCA = breast cancer susceptibility gene; b intake from wine was 0.0 of resveratrol for the 1st tertile, ranged between 0.1 and 176.8 for the 2 nd tertile, > 176.8 for the 3 rd tertile;  (Damianaki et al., 2000). In vitro experiments, polyphenols found in grapes showed the activity to induce cancer cells apoptosis and delay tumor growth (Castillo-Pichardo et al., 2009). In the animal model, transgenic mice also demonstrated decreased incidence rates of cancers with red wine solid food ingestion (Clifford et al., 1996). Secondly, some studies explored other mechanism pathways in which wine may serve as a kind of nutritional aromatase inhibitors (AI) (Byrne et al., 2002). The results showed that sex hormone binding globulin (SHBG) and luteinizing hormone (LH) were higher in serum with red wine consumption, which was explained by the hypothalamic up-regulation in response to lower estrogen levels. Thus, wine may not increase breast cancer risk via the hormonal shift patterns. Thirdly, a previous cohort study suggested that wine consumption induce breast density inversion in postmenopausal women after adjusting for other sources of alcohol (Boyd et al., 2006). Some epidemiologic studies have confirmed that breast density was risk factor for breast cancer (Flom et al., 2009). Other possible mechanisms of action need to be investigated in future.
In subgroup analysis by study design, case-control studies, especially hospital-based case-control studies, seemed to report much higher relative risks than cohort studies. The inconsistent findings may have been attributed to greater recall and selection biases in case-control studies because of their retrospective nature. And most non-high-quality studies are case-control ones, which further explain these results. When compared the RRs in different regions, we observed great difference in RR across geographic area. The RRs in European countries, especially France and Italy, were higher than that of USA and Canada. This may be due to the distinctions of diet patterns among different geographic regions. In many European countries, wine is usually an integral part of the resident's dietary habits daily diet.
Strengths of our studies include a large size (18106 breast cancer cases from 8 cohort studies and 18 case-control studies) and a quantitative dose-response analysis. Also, results from high-quality, cohort studies and studies adjusted for a variety of confounders are relatively consistent. Nevertheless, several limitations in our meta-analysis need to be mentioned. First of all, we noted that the majority of the cases were extracted from case-control studies, which are generally based on the memory and past record leading to more recall bias than cohort studies. Secondly, all the studies included only covered the Whites, lacking the diversity of races. Thirdly, as food-frequency questionnaires were used in each component studies, our findings were likely to be influenced by the underestimation of wine consumption. Besides, the potential misclassification of wine ingestion dose also may affect our results due to the broad range of definition of conversion in wine consumption.
Considering drinking is associated with increased risk of other health problems in women, such as birth defects, stroke, and other many types of cancers (Wiseman, 2008). Because wine consumption has increased in the general population, especially among young women, further research to clarify the relative safety in women is needed.
In conclusion, our analysis indicates that high dose of wine drinking is associated with increased risk of breast cancer, while low dose reduce the risk. However, future well-designed cohort or interventional studies are needed to confirm the findings and elucidate the underlying mechanisms.   I  I  I  -II  I  I  I  I  9 Martin-Moreno JM, 1993, Spain   I  I  I  I  II  I  I  I  I  10 Ferraroni M, 1991, Italy -I -I  I  I  I  -I  6 Rosenberg L, 1990, Canada -I . I  -II  I  I  I  -7 Toniolo P, 1989, Italy  I  I  I  -II  -I  I  I  8   Richardson  S, 1989,  France   -I  -I  I  -I  I  -5 Adami HO, 1988, Sweden and Norway   I  I  I  I  II  I  I  I  -9 La Vecchia C, 1985, Italy Talamini R, 1984, Italy - Lê MG, 1984, France - Paganini-Hill A, 1983, USA a A study could be awarded a maximum of one star for each item except for the item Control for important factor or additional factor. b A maximum of 2 stars could be awarded for this item. Studies that controlled for smoking and alcohol received one star, whereas studies that controlled for other important confounders such as family history or fresh vegetables and fruit intake received an additional star. c One star was assigned if there was no significant difference in the response rate between control subjects and cases by using the chi-square test (P>0.05)