Alcohol Consumption by Beverage Type and Risk of Breast Cancer

A Dose-Response Meta-Analysis of Prospective Cohort Studies

Qiuyu Sun; Weihong Xie; Yanli Wang; Feifei Chong; Mengmeng Song; Tiandong Li; Linping Xu; Chunhua Song


Alcohol Alcohol. 2020;55(3):246-253. 

In This Article


Search Strategy

Both Pubmed and Web of Science were used to search for epidemiological studies from 1990 to 2018 on alcohol intake and breast cancer incidence. The search included the (all fields) search terms 'breast cancer', 'incidence', 'risk' and 'alcohol intake' in different combinations (Supplementary Table S1) and resulted in a total count of 8025 articles after excluding duplicate articles. At the same time, we searched and identified five articles that are not captured by the two databases through reviewing reference lists in retrieved articles. Literature search and review process were enforced based on the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) (Moher et al., 2009) and Meta-analysis of Observational Studies in Epidemiology (MOOSE) guidelines (Stroup et al., 2000).

Study Eligibility

Some studies were included in the meta-analyses when they met the following strategies: (1) cohort studies published as original articles, (2) specific daily drinking dose and/or more detailed stratification, (3) outcome variable defined as the incidence of breast cancer and (4) relative risk (RR) or hazard ratio (HR) reported with corresponding 95% confidence intervals (CI) of breast cancer or provided relevant data to calculate. If the article was duplicated or from the same study population, only the study which is the most recent or the most complete was included (Land et al., 2014). Based on a review of the titles and abstracts of identified articles, we first excluded ecological studies and cross-sectional studies. After carefully reading the full text of the remaining literature, we further excluded the articles that met any of the following conditions: case reports or case series, no data on alcohol dose in the general population, no clinical breast cancer outcomes, or no outcome of morbidity (Figure 1).

Figure 1.

Flow chart of literature search for studies investigating association between alcohol consumption and breast cancer risk (search until 1 December 2018).

Data Abstraction

The Newcastle–Ottawa scale (NOS) Wells et al. (2013) includes three domains that were used to assess study quality and extract information on possible sources of heterogeneity by two authors (QY Sun and WH Xie) and then reviewed by a third investigator. For each study, descriptive information including population source, participant characteristics, exposure assessment, outcome ascertainment and adjustment for confounding factors were abstracted. For every exposure category, the number of cases, person-years (if incidence rate) or persons (if cumulative incidence), the measure of association (e.g. rate ratio, hazard ratio), a measure of statistical uncertainty of the measure of association (confidence interval), the specific dose values of each exposure category and the model adjusted with the most potential confusion were extracted. The alcohol exposure categories and measures of association from each study are presented in Supplementary Table S2.

Statistical Analysis

Pooled measure was calculated as the inverse variance weighted mean of the logarithm of RRs with 95%CI to assess the association of alcohol intake with breast cancer risk. The appropriateness of effect model was assessed by using a test of heterogeneity (χ2-test). In the absence of significant heterogeneity (P heterogeneity > 0.10 or I 2 < 50%), the fixed effects model (Mantel and Haenszel, 1959) was used to estimate the summary RRs and their 95%CI. If there was obvious heterogeneity (P heterogeneity < 0.10 or I 2 ≥ 50%), a random effects model (DerSimonian and Laird, 1986) was selected. To explore the differences in risk estimates between possible sources of heterogeneity, a subgroup analysis examined for study region, follow-up time, menopausal state, number of subjects and number of case. To assess whether there was evidence for small-study effects (i.e. whether smaller studies tend to give substantially larger estimates of effect size compared with larger studies) (Sterne et al., 2011), Egger's test was conducted (Egger et al., 1997). Sensitivity analysis was used to assess the robustness and potential heterogeneity of aggregated results by omitting a study at a time.

For the dose-response analysis, the assigned dose was determined as the exposure value in each category. Generalized least squares (GLS) regression methods were used to estimate dose-response trends. By way of use of the mvmeta command, method of moments (mm) are used to check whether the combined effect quantities conform to the nonlinear trend. If P < 0.05 a nonlinear spline model is used and, otherwise, a linear model. Finally, acquire the overall pooled risk estimates obtained by all included studies and the most recent estimates of proportions of alcohol in these populations.

The ARP of breast cancer for current alcohol consumption compared with never drinker was estimated for North America, Asia and Europe. Stata statistical software was used for all analyses (version 12; Release 12. College Station, TX: StataCorp LP).