Although all studies do not report an association between alcohol intake and breast cancer risk,1 there is accumulated evidence for a positive association from both cohort and case-control studies.2, 3, 4, 5, 6 A large collaborative reanalysis of individual data from 53 different epidemiologic studies found a dose–response effect of alcohol.7 However, because of reporting errors of alcohol intake, uncertainty remains about the quantitative effects of intake of a fixed amount of alcohol. Furthermore, firm conclusions about the risk of breast cancer at low intake levels of alcohol cannot be drawn because of the likelihood of measurement errors, particularly the tendency of underestimation.
Some of the earlier studies found a dose–response effect,2, 3, 4, 5, 8 while others found an effect at high intake levels only.6, 9, 10, 11 Several studies have examined the separate effects of beer, wine and spirits; but the type of alcoholic beverage does not appear to differentially influence the risk of breast cancer.1, 3, 5
One meta-analysis2 addressed several issues concerning the disparity in outcome from different studies. Risk estimates were higher in hospital-based case-control studies compared to cohort or community-based case-control studies; cohort studies with a follow-up of <10 years had higher estimates compared to cohort studies with longer follow-up. No differences were found depending on publication year, menopausal status and location of study (United States vs. outside United States). However, Longnecker3 found that studies with the strongest alcohol–breast cancer association came from countries with high per capita alcohol consumption, i.e., Western European countries. Studies indicate that different cultures could imply differences in alcohol drinking patterns, type of alcoholic beverage used and the association between alcohol intake and potential dietary (e.g., fruit and vegetable intakes) and nondietary confounders.12, 13
Only a limited number of studies have examined alcohol intake together with other dietary factors. The association between alcohol intake and breast cancer was independent of dietary habits in general in some studies.3, 6 However, other studies found that high folate consumption might reduce breast cancer risk at high alcohol intake levels.14, 15 Yet another study indicated higher risks in strata of higher absolute fat intake.9 The case of adjusting for fat intake is of special analytic interest since both alcohol and fat potentially influence breast cancer risk, though the role of fat is more controversial16 than that of alcohol. The common approach in analytic projects is to examine total fat intake as relative intake based on TE. Since alcohol, like fat, is an energy source, this definition leads to an inherent inverse correlation between fat and alcohol intake. However, this is true only if alcohol and fat intakes are independent variables. If this is not the case and, e.g., a positive correlation exists, the inverse correlation between relative intakes will be influenced and weakened. Furthermore, current nutrient recommendations concerning macronutrient composition of the diet express macronutrient intakes as percentages of NAE, not of TE.17 It is thus of interest to evaluate the influence of the 2 definitions of energy when estimating breast cancer risk associated with alcohol, with simultaneous adjustment for fat.
Our aims were to examine (i) if breast cancer risk in postmenopausal women is associated with intakes of total alcohol, specific alcoholic beverages or total fat; (ii) the genuine effects of total alcohol and fat intakes when adjusted for each other; and (iii) if the specific energy-adjustment approach influences these associations.
MATERIAL AND METHODS
The MDC study is a prospective cohort study in Malmö, a city in the south of Sweden with approximately 250,000 inhabitants. The MDC source population was, in 1991, defined as all persons living in the city of Malmö and born during 1926–1945. However, in May 1995, the cohort was extended to include all women born during1923–1950 and all men born during 1923–1945. With this extension, 74,138 persons constituted the source population. Inadequate Swedish language skills and mental incapacity were the only exclusion criteria. When baseline examinations closed in October 1996, 28,098 participants had completed all parts. Details of recruitment and the cohort are described elsewhere.18 The ethics committee at Lund University has approved the MDC study protocol. Participants signed informed consent.
Participants visited the MDC screening center twice. During the first visit, they were informed about the study and instructed how to register meals in the menu book and how to fill in the diet questionnaire. Project nurses took blood samples, blood pressure and anthropometric measurements. Participants completed all questionnaires at home. During the second visit, approximately 2 weeks after the first, the socioeconomic questionnaire was checked for completeness and the dietary interview conducted.
We used an age-based definition of menopausal status.19 The questionnaire item “What year did your menses cease?” was answered by all participants, but it was unclear to the women how to relate the answer to surgery (hysterectomy or oophorectomy) or hormone therapy use. Further, the computerized information on hysterectomy and oophorectomy from Malmö University Hospital does not cover the entire relevant time period for the whole MDC cohort. Natural menopause was determined in a group of women (n = 2,898) in the MDC cohort without surgery and without hormone therapy. The median natural menopause age was 50.0 years.
Eligible participants were thus women 50 years or older at baseline examination. All prevalent cancer cases, except cervical cancer in situ and nonmalignant melanoma skin cancer, were excluded. In total, 11,726 postmenopausal women were included in this subsample. The average follow-up time was 7.6 years. Women born during 1923–1925 and 1946 have shorter follow-up due to the MDC study design.
Case definition and ascertainment
The National Swedish Cancer Registry provided data until December 1999. Additional information until end of follow up (31 December 2001) was obtained from the Southern Swedish Regional Tumor Registry. Cases were women diagnosed during follow-up with invasive breast cancer or breast cancer in situ. Information on vital status was obtained from the National Tax Board, which provides up-to-date information on vital status for all Swedish residents. Cases contributed person-time from date of enrollment until time of diagnosis. Noncases contributed person-time from date of enrollment until death (470 women), migration from Sweden (57 women) or end of follow-up (31 December 2001), whichever was the first. In total, 342 incident (312 invasive and 30 in situ) breast cancer cases were documented during 89,602 person-years of follow-up.
The MDC uses an interview-based, modified diet history method. It combines (i) a 7-day menu book for registration of lunch and dinner meals, including cold beverages and alcohol, and (ii) a questionnaire for assessment of meal patterns, consumption frequencies and portion sizes of regularly eaten foods (i.e., sandwiches, cakes and cookies, fruit, breakfast cereals, milk and yoghurt, coffee and tea, sweets, snacks). Drugs, natural remedies and nutrient supplements were recorded in the menu book. During the interview, participants were asked complementary questions on their usual meal pattern, cooking methods and details of food choices. The participant at home used a booklet with 48 black-and-white photographs to estimate portion sizes. Usual portion sizes of foods and dishes listed in the menu book were estimated during the interview from a more extensive book with black-and-white photographs. Diet interviewers coded and entered the information from the menu book during the interview, using interactive software (Kostsvar; Aivo, Stockholm, Sweden).
Mean daily intake of foods was calculated based on frequency and portion size estimates from the questionnaire and menu book. The food intake was converted to energy and nutrient intakes using the MDC nutrient database, where the majority of the nutrient information comes from the Food Composition Database, Version PC-K0572-93, from the National Food Administration (Uppsala, Sweden).
The relative validity of the MDC method was evaluated in a sample of Malmö residents, 105 women and 101 men, 50–69 years old, using weighed food records, 3 days every second month during a year (i.e., 18 daily food records), as the reference method. Pearson correlation coefficients between the reference method and the MDC method administrated after the 12-month reference period were 0.69 for energy-adjusted fat intake, 0.78 for energy-adjusted total alcohol,20 0.77 for beer, 0.65 for wine and 0.71 for spirits.21 Reproducibility for total alcohol and alcoholic beverages was high.22
Daily intakes of TE (kcal/day), NAE (kcal/day), fat (g/day), E% from fat, total alcohol (g/day), E% from alcohol, beer (i.e., beer with 1.8, 2.8 or 4.5% alcohol and cider with 0.8% alcohol; cl/day), wine (aggregated code including white, red and fortified wines; cl/day), spirits (cl/day), the sum of folic acid (μg/day) from diet and dietary supplements and the sum of all fruits and vegetables (g/day) were calculated from the dietary information. The contribution of alcohol from each beverage group, beer-alcohol, wine-alcohol and spirits-alcohol, was calculated as the percentage contribution of the total amount of alcohol ingested from all alcoholic beverages.
Two relative fat intake variables were defined: FATTE and FATNAE. For each definition, 5 exposure categories were created based on the quintile ranking of participants on fat residuals.
Information on total alcohol consumption was converted into a 4-category variable. Women reporting 0 consumption in the menu book and no consumption of any type of alcohol during the previous year in the questionnaire were categorized as total abstainers. The other women were categorized according to an assumption of biologic risk;23 this definition has previously been used in several projects from the MDC cohort.24 Category ranges were <15 g alcohol/day (low), 15–30 g (medium) and >30 g (high).
Intakes of beer, wine and spirits were categorized into 4-category variables based on the ranking of absolute intakes of each beverage: abstainers (0 consumption in menu book and no consumption of any type of alcohol during the last year in the questionnaire); low consumers (intakes below or equal to the median); medium consumers (intakes above median but below or equal to the 97.5th percentile); high consumers (intakes above the 97.5th percentile).
In September 1994, the processing of dietary data was slightly altered.25 Method version (indicating data collection before or after 1 September 1994), diet interviewer and season of diet interview were used to control for undue variation in dietary data collection over time.
Age at baseline was obtained from the 10-digit personal identification number.
Information on reproduction, socioeconomic factors and lifestyle factors was collected by a structured multiple-choice questionnaire.
Past change of dietary habits was based on the questionnaire item “Have you substantially changed your dietary habits because of illness or other reason?”
Age at menarche was used as a continuous variable. Age at birth of first child was divided into 4 categories (≤24 years, >24–30 years, >30 years and “no children”), missing information (2.0%) was recoded as “no children”. Current use of hormone therapy (dichotomous variable, yes/no) was based on the questionnaire item “Which medications do you use on a regular basis?” in combination with information on drug use from the 7-day menu book.26
Participants were divided into 4 categories according to highest level of education: ≤8 years, 9–10 years, 11–12 years and college education/university degree.
Participants indicated the number of minutes per week spent on 17 different physical activities, separately for the 4 seasons.27 Based on this information, an overall leisure time physical activity score was created.28 The score was divided into quintiles and further categorized as low (quintile 1), moderate (quintiles 2–4) or high (quintile 5). Participants were categorized according to smoking habits as current smokers (including irregular smokers), ex-smokers or never-smokers.
Standing height was measured with a fixed stadiometer calibrated in centimeters. Weight was measured to the nearest 0.1 kg using a balance-beam scale with subjects wearing light clothing and no shoes. Waist was measured midway between the lowest rib margin and iliac crest.
To evaluate low energy reporting, which is a major concern in dietary assessment,29, 30, 31, 32 the ratio between reported total EI and BMR was calculated. BMR was calculated using the equation recommended by the WHO, based on age, sex, weight and height.33
All dietary variables were log-transformed to normalize distributions. The SPSS (Chicago, IL) statistical computer package (version 10.0) was used for statistical analyses.
Both fat and alcohol contribute to total EI. Different models for energy adjustment are discussed in the literature.34, 35, 36, 37 We compared 2 different energy-adjustment models, adjusting for TE or NAE:
In model (1), β1t symbolizes the change in IRRt with increased alcohol intake, while FATTE and TE are constant, i.e., a substitution of alcohol energy for carbohydrate or protein energy. β2t symbolizes an increase in risk with increased FATTE when alcohol and TE intakes are constant, i.e., a substitution of fat energy for carbohydrate or protein energy. In model (2), β1n symbolizes the change in IRRn with increased alcohol intake, while FATNAE and NAE are constant, i.e., adding alcohol energy. β2n symbolizes an increase in risk with increased FATNAE when alcohol and NAE are constant, i.e., a substitution of fat energy for carbohydrate or protein energy.
The bivariate relation between disease status and exposure variables and covariates was examined with Student's t-test and χ2 test. Bivariate relations between categories of FATTE; FATNAE; wine, beer, spirits and total alcohol intakes; and covariates were examined with ANOVA and χ2 test. The examined variables were diet interviewer, season of diet interview, past food habit change, method version, age at baseline, height, waist, current hormone use, age at birth of first child, age at menarche, leisure time physical activity, smoking habits, educational level, EI:BMR, E% fat, E% alcohol, beer-alcohol, wine-alcohol, spirits-alcohol, folic acid and total fruit and vegetables.
Cox regression examined the associations between intakes of beer, wine, spirits, total alcohol and relative fat and breast cancer incidence. Low consumers of total alcohol and alcoholic beverages were chosen as the reference category since abstainers probably are heterogeneous and may not represent the lowest risk. Further, the abstainer categories had fewer person-years of follow-up. First, the influences of beer, wine, spirits, total alcohol and relative fat intakes on breast cancer risk were examined in separate models while controlling for diet interviewer, season of diet interview, method version, age at baseline, past food habit change and energy. In a second step, the multivariate model was extended to include both known nondietary risk factors38 and potential confounders, i.e., height, waist, current hormone use, age at birth of first child, age at menarche, leisure time physical activity, smoking habits and educational status. The full models were also repeated while adjusting for folic acid intake and total fruit and vegetable intake.
In addition, the genuine effects of total alcohol and relative fat intakes on breast cancer risk were examined in mutually adjusted models, both the basic and the full models.
Finally, the full models were repeated with exclusion of women younger than 51 years at baseline (n = 719, including 27 cases), cases diagnosed with incident breast cancer within 1 year from baseline examination (n = 31) or cases with noninvasive breast cancer (n = 30).
Cases were younger at baseline (a consequence of the MDC study design and shorter follow-up among many older women). In addition, cases were taller, were more often current users of hormone therapy and had more often medium level of physical activity and high wine intake (Table I). Height, waist and current hormone use were significant risk factors for breast cancer in all multivariate models (data not shown).
Table I. Baseline Characteristics, Means (SD) or Proportions (%) by Breast Cancer Incidence Status in the MDC Cohort 1991–2001
|Age (years)||60.4 (6.5)||59.4 (5.8)||<0.01|
|Height (cm)||163.0 (5.9)||164.2 (6.0)||<0.01|
|Waist (cm)||78.5 (10.6)||79.0 (10.5)||0.42|
|Age at menarche (years)||13.7 (1.5)||13.6 (1.4)||0.54|
|Total vegetable and fruit (g/day)||453 (189)||452 (173)||0.93|
|Folic acid (μg/day)||231 (72)||234 (73)||0.38|
|EI:BMR||1.45 (0.39)||1.46 (0.38)||0.90|
|E% fat2||37.4 (5.9)||37.9 (5.8)||0.20|
|E% alcohol3||1.45 (1.7)||1.48 (1.8)||0.78|
|Current hormone therapy (%)|| || || |
| No||79.9||68.1|| |
|Age at first child (%)|| || || |
| ≤24 years||44.3||42.1||0.74|
| >24 to ≤30 years||32.4||32.2|| |
| >30 years||9.0||10.2|| |
| No children (includes missing)||14.2||15.5|| |
|Leisure time physical activity (%)|| || || |
| Medium||59.8||67.8|| |
| High||20.1||17.1|| |
|Smoking status (%)|| || || |
| Ex-smoker||26.5||29.8|| |
| Never-smoker||47.6||43.6|| |
|Education level (%)|| || || |
| ≤8 years||45.8||41.1||0.28|
| 9–10 years||30.4||33.4|| |
| 11–12 years||12.9||12.6|| |
| University degree||10.9||12.9|| |
|Beer intake (%)|| || || |
| ≤5.6 cl/day||45.2||44.4|| |
| >5.6 to ≤36.1 cl/day||43.5||45.9|| |
| >36.1 cl/day||2.3||2.9|| |
|Wine intake (%)|| || || |
| ≤2.9 cl/day||44.6||44.4|| |
| >2.9 to ≤20.8 cl/day||40.9||38.3|| |
| ≥20.8 cl/day||2.1||4.7|| |
|Spirits intake (%)|| || || |
| 0 cl/day||62.7||67.0|| |
| >0 to ≤2.5 cl/day||22.1||17.8|| |
| >2.5 cl/day||2.3||2.6|| |
|Total alcohol intake (%)|| || || |
| ≤15 g/day||77.1||78.7|| |
| >15 to ≤30 g/day||12.3||11.4|| |
| >30 g/day||1.9||3.2|| |
|FATTE (%)|| || || |
| Quintile 1||20.0||18.4||0.12|
| Quintile 2||20.1||15.2|| |
| Quintile 3||19.9||23.5|| |
| Quintile 4||20.0||21.1|| |
| Quintile 5||19.9||22.2|| |
|FATNAE (%)|| || || |
| Quintile 1||20.1||17.8||0.63|
| Quintile 2||20.1||18.1|| |
| Quintile 3||20.0||20.5|| |
| Quintile 4||19.9||21.6|| |
| Quintile 5||19.9||21.9|| |
Covariates differed across total alcohol exposure categories. For instance, high consumers of alcohol were younger, were taller, ate less fruit, had higher EI:BMR ratios (i.e., were less likely to report low energy), more often used hormone therapy, were younger at birth of the first child, had higher leisure time physical activity, were more often smokers and had longer education compared to those in the other alcohol intake categories (Table II). Differences across categories of relative fat intake were small and similar for both definitions (FATTE and FATNAE). Briefly, a larger proportion of low fat consumers were never-smokers, had high leisure time physical activity and had changed dietary habits in the past. There were only small or no differences in the proportions of current hormone therapy use, educational level or age at birth of the first child. Mean differences were small or nonexistent for age at baseline, waist, height, age at menarche and EI:BMR ratio. Low fat consumers had higher intakes of total fruit and vegetables and of folate. There were no differences in alcohol source across quintiles of FATTE; however, wine and spirits contributed more alcohol and beer less alcohol in FATNAE quintiles 4 and 5 (data not shown).
Table II. Baseline Characteristics, Means (SD) or Proportions (%), Across Categories of Total Alcohol Intake in Women with no Diagnosed Breast Cancer in the MDC Cohort 1991–2001
|Age (years)||62.2 (6.3)||60.5 (6.5)||58.6 (6.4)||57.6 (6.3)||<0.001|
|Height (cm)||161.0 (6.1)||163.0 (5.9)||164.6 (5.68)||165.0 (5.9)||<0.001|
|Waist (cm)||81.9 (12.5)||78.4 (10.5)||76.6 (9.4)||79.3 (10.2)||<0.001|
|Age at menarche (years)||13.8 (1.6)||13.7 (1.5)||13.6 (1.4)||13.6 (1.7)||0.009|
|Total fruit and vegetables (g/day)||463 (214)||458 (190)||443 (183)||405 (195)||<0.001|
|Folic acid (μg/day)||223 (79)||232 (72)||234 (71)||223 (74)||<0.001|
|EI:BMR||1.38 (0.44)||1.44 (0.38)||1.54 (0.37)||1.55 (0.34)||<0.001|
|E% fat2||37.0 (6.9)||37.5 (5.8)||37.5 (5.4)||35.4 (5.5)||<0.001|
|E% alcohol3||—||1.0 (1.0)||4.0 (1.3)||7.6 (2.4)||<0.001|
|Beer-alcohol as percentage of total alcohol intake||—||47.0 (40.1)||18.2 (19.0)||17.4 (20.8)||<0.001|
|Wine-alcohol as percentage of total alcohol intake||—||43.8 (40.0)||67.6 (25.2)||63.9 (25.9)||<0.001|
|Spirits-alcohol as percentage of total alcohol intake||—||9.2 (20.6)||14.2 (18.2)||18.7 (19.3)||<0.001|
|Current hormone therapy? (%)|| || || || ||<0.001|
| Yes||12.5||19.1||30.0||29.1|| |
| No||87.5||80.9||70.0||69.1|| |
|Changed dietary habits? (%)|| || || || ||<0.001|
| Yes||35.7||26.4||17.7||24.5|| |
| No||64.3||73.6||82.3||75.5|| |
|Age at first child (%)|| || || || ||<0.001|
| ≤24 years||50.2||44.1||40.9||48.6|| |
| >24 to ≤30 years||24.7||33.1||33.3||31.8|| |
| >30 years||9.7||9.0||9.0||7.7|| |
| No children (includes missing)||15.4||13.8||16.8||11.8|| |
|Leisure time physical activity (%)|| || || || ||<0.001|
| Low||29.7||19.6||15.9||20.1|| |
| Medium||53.1||60.5||62.4||56.8|| |
| High||17.2||19.9||21.7||23.1|| |
|Smoking (%)|| || || || ||<0.001|
| Yes (includes irregular)||26.7||24.6||31.0||42.3|| |
| Ex-smoker||17.0||26.2||33.8||32.7|| |
| Never-smoker||56.3||49.2||35.2||25.0|| |
|Education (%)|| || || || ||<0.001|
| ≤8 years||67.6||47.1||26.7||17.7|| |
| 9–10 years||19.3||30.6||35.8||38.6|| |
| 11–12||8.5||12.2||19.2||22.3|| |
| University degree||4.6||10.1||18.3||21.4|| |
Cross-classification of the total alcohol and relative fat intake categories showed different patterns for FATTE and FATNAE, respectively (Table III). Medium and high consumers of total alcohol were more often classified in the lower FATTE quintiles but in the highest FATNAE quintile. Women classified as low alcohol consumers were evenly distributed across quintiles of both relative fat intake definitions.
Table III. Cross-Classification (Number and Proportions) of Alcohol Intake Categories Across Quintiles of Relative Fat Intake in Women with no Diagnosed Breast Cancer in the MDC Cohort 1991–2001
|FATTE|| || || || || || ||<0.001|
| Quintile 1||65 (20)||29.6 (3.2)||247 (25.0)||1 677 (19.1)||287 (20.4)||71 (32.3)|| |
| Quintile 2||78 (23)||34.6 (1.8)||178 (18.0)||1 777 (20.3)||289 (20.6)||49 (22.3)|| |
| Quintile 3||84 (25)||37.5 (1.8)||150 (15.2)||1 752 (20.0)||312 (22.2)||53 (24.1)|| |
| Quintile 4||90 (27)||40.3 (2.0)||170 (17.2)||1 770 (20.2)||300 (21.4)||33 (15.0)|| |
| Quintile 5||100 (31)||45.4 (3.5)||242 (24.6)||1 796 (20.4)||217 (15.4)||14 (6.4)|| |
|FATNAE|| || || || || || ||<0.001|
| Quintile 1||65 (21)||29.7 (3.2)||294 (29.8)||1 793 (20.4)||174 (12.4)||23 (10.5)|| |
| Quintile 2||77 (23)||34.6 (2.0)||192 (19.5)||1 860 (21.2)||210 (14.9)||21 (9.5)|| |
| Quintile 3||84 (25)||37.5 (2.2)||160 (16.2)||1 771 (20.2)||294 (20.9)||51 (23.2)|| |
| Quintile 4||91 (26)||40.4 (2.2)||165 (16.7)||1 733 (19.8)||325 (23.1)||48 (21.8)|| |
| Quintile 5||100 (31)||45.1 (3.9)||176 (17.8)||1 615 (18.4)||402 (28.6)||77 (35.0)|| |
High beer intakes were associated with a nonsignificantly elevated risk of breast cancer. High intakes of wine (>20.8 cl/day) significantly increased the risk for breast cancer compared to low intakes. Medium intakes of spirits (>0 to ≤2.5 cl/day) were associated with decreased risk of breast cancer. Both associations remained significant after adjustment for established risk factors (Table IV). Nonsignificantly elevated risks were found for high (>30 g/day) intake of total alcohol, both with and without adjustment for established risk factors (Tables V, VI). Results did not change substantially when we adjusted for folic acid or total fruit and vegetable intakes or excluded women younger than 51 years at baseline, cases diagnosed within 1 year from baseline or cases with in situ breast cancer (data not shown).
Table IV. IRR of Breast Cancer According to Intake Categories of Beer, Wine and Spirits in the MDC Cohort 1991–2001
|Beer (cl/day)|| || || || || || || || || |
| ≤5.6||3.0||40,363||152||1|| ||39,238||142||1|| |
| >5.6 to ≤36.1||7.4||39,091||157||1.08||0.86–1.35||38,123||155||1.11||0.88–1.40|
|Wine (cl/day)|| || || || || || || || || |
| ≤2.9||1.7||39,908||152||1|| ||38,838||144||1|| |
| >2.9 to ≤20.8||10.8||36,679||131||0.93||0.74–1.18||35,828||127||0.88||0.69–1.13|
|Spirits (cl/day)|| || || || || || || || || |
| 0||3.9||56,053||229||1|| ||54,616||218||1|| |
| >0 to ≤2.5||11.2||19,967||61||0.74||0.56–0.99||19,471||60||0.73||0.55–0.98|
Table V. IRR of Breast Cancer According to Intake Categories of Total Alcohol and Fatte in the MDC Cohort 1991–2001
|Total alcohol (g/day)|| || || || || || || || || || || || |
| ≤15||69,035||269||1|| ||1|| ||67,244||257||1|| ||1|| |
| >15 to ≤30||10,953||39||0.91||0.65–1.28||0.92||0.66–1.29||10,663||39||0.88||0.62–1.24||0.88||0.63–1.25|
|FATTE|| || || || || || || || || || || || |
| Quintile 1||18,210||63||1|| ||1|| ||17,745||61||1|| ||1|| |
| Quintile 2||18,045||52||0.84||0.58–1.22||0.84||0.58–1.22||17,654||48||0.80||0.55–1.18||0.81||0.55–1.18|
| Quintile 3||17,764||79||1.34||0.96–1.87||1.34||0.96–1.88||17,315||78||1.36||0.97–1.91||1.36||0.97–1.92|
| Quintile 4||17,745||72||1.23||0.87–1.72||1.24||0.90–1.74||17,242||70||1.25||0.88–1.77||1.26||0.89–1.79|
| Quintile 5||17,452||76||1.35||0.96–1.89||1.37||0.97–1.92||16,818||72||1.34||0.94–1.90||1.36||0.96–1.94|
| || || ||p for trend = 0.0183||p for trend = 0.0153|| || ||p for trend = 0.0183||p for trend = 0.0163|
Table VI. IRR of Breast Cancer According to Intake Categories of Total Alcohol and Fatnae in the MDC cohort 1991–2001
|Total alcohol (g/day)|| || || || || || || || || || || || |
| ≤15||69,035||269||1|| ||1|| ||67,244||257||1|| ||1|| |
| >15 to ≤30||10,953||39||0.94||0.65–1.28||0.88||0.63–1.24||10,663||39||0.88||0.62–1.24||0.85||0.60–1.20|
|FATNAE|| || || || || || || || || || || || |
| Quintile 1||18,219||61||1|| ||1|| ||17,774||58||1|| ||1|| |
| Quintile 2||18,016||62||1.04||0.73–1.48||1.03||0.72–1.47||17,570||59||1.05||0.73–1.51||1.05||0.73–1.51|
| Quintile 3||17,765||70||1.23||0.87–1.74||1.21||0.86–1.72||17,236||67||1.24||0.87–1.77||1.23||0.86–1.76|
| Quintile 4||17,725||74||1.31||0.93–1.84||1.30||0.92–1.83||17,280||74||1.38||0.97–1.96||1.38||0.97–1.96|
| Quintile 5||17,492||75||1.36||0.97–1.92||1.35||0.95–1.90||16,915||71||1.35||0.94–1.93||1.35||0.94–1.93|
| || || ||p for trend = 0.0353||p for trend = 0.0473|| || ||p for trend = 0.0433||p for trend = 0.0503|
Significant positive trends were found across quintiles of both relative fat intake variables when examined in separate models (Tables V, VI). Adjustment for total alcohol did not affect trends across FATTE quintiles (Table V), but the p values for trends across FATNAE quintiles became higher (Table VI). In the mutually adjusted models, the risk associated with high alcohol intakes was somewhat higher in the model testing substitution of alcohol energy for energy from carbohydrate or protein (FATTE model, Table V) compared to the models without adjustment for relative fat. However, the risk estimates in the model testing adding alcohol energy (FATNAE model, Table VI) were somewhat lower compared to the models without relative fat adjustment. Excluding women younger than 51 years at baseline or cases diagnosed within 1 year from baseline did not change results appreciably (data not shown). However, when in situ cancer cases were excluded, the p value for trend across fat quintiles became lower for FATTE (p = 0.004) and for FATNAE (p = 0.015). In addition, risk estimates were significantly increased in FATTE quintile 5 (IRR = 1.57, 95% CI 1.08–2.28) and in FATNAE quintiles 4 (IRR = 1.49, 95% CI 1.03–2.16) and 5 (IRR = 1.51, 95% CI 1.03–2.20).
Additional ad hoc analyses were performed with the model estimating the genuine risk of wine. Adjusting for total alcohol, beer or spirit intakes (continuous) in separate models did not affect risk estimates for wine. Adjustment for total alcohol intake categories did not affect risk estimates for wine (IRR = 2.38, 95% CI 1.14–5.06), but the nonsignificant elevated risk among high consumers of total alcohol did not remain (IRR = 0.92, 95% CI 0.40–2.15).
High consumption of total alcohol was associated with a nonsignificantly elevated risk of breast cancer. High intake of wine was associated with significantly increased breast cancer risk. We found a significant positive trend in risk across total fat intake quintiles. In the mutually adjusted models, the risk associated with substitution of alcohol energy (FATTE) had somewhat higher estimates compared to the model adding alcohol energy (FATNAE). Mutual adjustment did not affect the significant trend across fat quintiles in the FATTE model, while the p value for trend increased in the FATNAE model.
Observational epidemiologic studies in general show an association between alcohol intake and breast cancer risk.5, 6, 39 Also, there are several hypotheses concerning the biologic effect of alcohol on breast cancer development: an influence on estrogen levels through decreased steroid metabolism,40 an influence on the production of reactive oxygen species via alcohol metabolism41 and inhibition of DNA repair and greater metastatic potential of breast cancer cells.42 It has also been suggested that the alcohol dehydrogenase-3 genotype may modify the alcohol effect.43
Our results support a threshold effect; we have no indication of elevated risk for reported total alcohol consumption <30 g/day, corresponding to approximately 2 glasses (30 cl) of wine daily. However, due to misreporting of alcohol, the absolute level of alcohol intake required for an effect remains unclear. Self-reports of alcohol intake must be used with caution when trying to establish a level where alcohol exerts a biologic effect. In our study, intakes of alcoholic beverages were recorded in the menu book during 7 consecutive days. The relative validity and reproducibility of the MDC diet history for total alcohol and alcoholic beverages are high.20, 21, 22 Nevertheless, the reported absolute intakes of total alcohol were, in women, substantially lower than with the reference method.20 Food record data on alcohol intake has been found to better predict the association between alcohol intake and estradiol levels compared to self-reported usual consumption.44 Information on alcohol habits is sensitive, prone to reporting bias; it is possible that women in this cohort reported wine more accurately than beer and spirits. This could mean higher misclassification of women into total alcohol intake categories compared to wine intake categories, which might explain the nonsignificant association between total alcohol and breast cancer risk. In this population, the correlation between total alcohol and alcohol from wine is high (0.88) compared to the correlations with alcohol from beer (0.48) or alcohol from spirits (0.55). In addition, as shown in Table IV, median intakes of total alcohol differed between the different alcoholic beverages. “High wine consumption” was the only category with a median total alcohol intake >30 g/day. Thus, because wine is the major source of alcohol in this population, it is difficult to separate the effect of wine from that of total alcohol. In the model including both wine and total alcohol intake categories, only the elevated risk associated with high wine intake remained.
It is difficult to elucidate the genuine effect of alcohol from a web of surrounding confounders (Table II) in observational epidemiologic studies. We adjusted for known risk factors of breast cancer and selected nutritional variables, but there might be factors not accounted for, e.g., psychologic and personality factors45 or other nutritional factors, that influence results. In addition, there is a risk of residual confounding. For instance, the influence of higher vs. lower education, which is interrelated with a number of factors, including social class, is difficult to capture with one question. Thus, even when models are adjusted for education, a residual effect might remain.
Few associations are so extensively studied in epidemiology as the one between total fat and breast cancer risk. Measurement errors, different energy-adjustment models and the diversity and range of fat intakes might be factors contributing to the conflicting results observed in epidemiologic studies.16, 46, 47, 48, 49, 50, 51, 52, 53, 54
One plausible role of dietary fat is via an influence on the levels of bioavailable estrogen. A meta-analysis of 13 intervention studies showed that a reduction of fat intake significantly reduced serum estradiol levels.55 The authors concluded that reducing fat intake below 20% of energy might influence breast cancer risk. However, they also pointed out that it is difficult to elucidate the genuine effect of fat per se. Weight loss and increased fiber intake could have influenced the results in the intervention studies.
One hypothesis suggests overeating of total, especially saturated, fat as the starting point of a series of events finally leading to a cycle of insulin resistance, increased plasma levels of insulin, decreased levels of sex hormone–binding globulin, increased levels of free estrogen and increased ovarian androgen production. This leads to a hormonal–metabolic profile that is associated with an increased risk of breast cancer.56
In animal studies, the type and amount of fat are important determinants of the effect of fat. The promoting effect appears to be dependent on the ratio of n-6 to n-3 fatty acids in the diet. Eicosapentaenoic acid (n-3) especially lacks tumor-promoting effects in animal models. The mechanism is believed to depend on the eicosanoids formed. Eicosapentaenoic acid inhibits eicosanoid production from the n-6 family by competing for desaturates.57 Besides the influence on eicosanoid production, other mechanisms are also proposed for the effect of fatty acids, e.g., DNA damage caused by reactive oxygen species from polyunsaturated fatty acids, interaction between fatty acids and genomic DNA leading to alterations in gene expression, influences on hormone levels, structural and functional changes in cell membranes influencing hormone and growth factor receptors and effects on immune function.57 Only a few epidemiologic studies have examined the effect of different fatty acids. Observations in the MDC cohort58 suggest that high intakes of n-6 fatty acids increase the risk of postmenopausal breast cancer when mutually adjusted for other types of fat. A few other epidemiologic studies have reported similar findings.52, 59, 60
Contrary to many prospective studies, our results support an effect of total fat independent of EI. Interestingly, the association became stronger when in situ cancer cases were excluded. We used relative fat intake, FATTE and FATNAE, as fat exposure variables. Relative fat intake is a complex variable, including aspects of the total diet and with varying associations with other nutrients and food groups. For instance, plant foods show strong inverse associations to relative fat intake in the MDC population.61 Food sources of fat and fatty acid composition differ across populations.62, 63, 64 Thus, total fat intake may not be regarded as the same exposure in different populations, which makes comparisons across studies difficult. The high validity20 in the MDC study could contribute to the ability to detect associations not found in other studies.58
Energy adjustment: variables and models
The bivariate association between relative fat and total alcohol intake differed depending on the energy definition used for the calculation of relative fat (Table III). The FATNAE model could be interpreted as the association between food energy and alcohol: e.g., subjects who consume a great deal of energy from fatty foods also drink a great deal of alcohol. The FATTE model illustrates the association between fat energy and alcohol energy relative to total EI. The crude correlation between absolute fat and alcohol intake is low (r = 0.073). The FATTE model introduces an inverse association between fat and alcohol by the way relative fat is expressed.
In contrast, outcomes of the multivariate analysis differed only slightly for the 2 relative fat definitions. Risk estimates in multivariate models had similar directions, though the p for trend was lower for FATTE compared to FATNAE. This indicates that these variables may be interchangeable when estimating risks. In addition, both models test substitution of fat energy for carbohydrate or protein energy; thus, the research question is the same. The risk estimates for total alcohol, however, represent 2 different hypotheses: exchanging alcohol when TE is constant or adding alcohol energy (i.e., increasing TE). The FATTE model could be regarded as a question of the ethanol effect per se. The FATNAE model, however, estimates an effect of both increasing ethanol and increasing energy. In this population, the first model yielded somewhat higher risk estimates. The differences in estimates are, however, small between the models. One reason could be the low mean reported total alcohol intake (1.5 E%). The outcomes of the 2 models, i.e., TE vs. NAE, might be different in populations with higher alcohol intakes.
Bivariate analyses indicated a lower EI:BMR ratio among alcohol abstainers and low consumers. If participants with “true” high total alcohol intakes misreported their alcohol intakes and were classified as abstainers or as having low total alcohol intakes, the risk associated with high consumption of total alcohol might be underestimated. In contrast, the differences in EI:BMR ratios across FATTE and FATNAE quintiles were small, mean ratios ranging 1.48–1.52. Thus, it is less likely that the observed positive trend in breast cancer risk across fat intake quintiles is influenced by differential underreporting of energy among low and high fat consumers.
To conclude, our results support the hypothesis of a threshold effect of alcohol intake on breast cancer risk. A nonsignificantly elevated breast cancer risk was found among women reporting high (>30 g/day) total alcohol intake. The major source of alcohol was wine, and high consumption of wine was significantly associated with increased breast cancer risk. We found a significant linear trend in increased risk across quintiles of relative fat. Mutual adjustment did not substantially affect risk estimates for total alcohol or relative fat intakes. The specific energy-adjustment approach (i.e., adjusting for TE or NAE) did not influence associations differentially.