SEARCH

SEARCH BY CITATION

Keywords:

  • breast cancer;
  • milk consumption;
  • diet;
  • cohort study

Abstract

  1. Top of page
  2. Abstract
  3. MATERIAL AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. REFERENCES

Analyses of dairy consumption and breast cancer incidence have yielded conflicting results. In this prospective cohort study of 48,844 premenopausal Norwegian women, we examined the relationship between childhood and adult milk consumption and breast cancer incidence. During a mean follow-up time of 6.2 years, 317 incident cases of breast cancer were diagnosed. Information on childhood and adult milk consumption was obtained from frequency questions mailed to the participants in 1991–92. Milk consumption as a child was negatively associated with subsequent breast cancer among the youngest women (34–39 years) (p for trend = 0.001), but not among older ones (40–49 years). Adult milk consumption tended to be negatively related to breast cancer incidence (p for trend = 0.12) after adjustment for age, reproductive and hormonal factors, body mass index, education, physical activity, and alcohol consumption. Women drinking more than 3 glasses of milk per day had an incidence rate ratio of breast cancer of 0.56 (95% confidence interval 0.31–1.01) compared with women not drinking milk. Analyses according to type of milk consumed and milk fat consumption did not reveal any clear associations. A combination of childhood and adult milk consumption produced a clear negative trend in breast cancer incidence rate ratios with increasing milk consumption (p = 0.03). © 2001 Wiley-Liss, Inc.

Breast cancer is the most common female cancer in the world and has an age-adjusted incident rate of 71.4 per 100,000 in Norway.1 Despite a constant search for knowledge about the biological mechanism operating behind the disease and means of preventing the disease, much is unknown and no efficient preventive methods have been revealed. Hormonal factors and reproductive history are known to influence breast cancer risk,2, 3 but hardly have preventive potential. Diet, on the other hand, is modifiable, but the relation to breast cancer is not as consistent.

Some of the strongest indications that diet plays a role in the etiology of breast cancer have emerged from ecological and migration studies. Several of the migrant studies have demonstrated that the breast cancer incidence may not be substantially changed for the immigrants themselves but rather for the second- and third-generation immigrants.4 This delay may indicate a stronger influence of childhood lifestyle factors, e.g., diet, than adult lifestyle factors on breast cancer risk, and harmonizes with a long latency period for breast cancer. Also, the positive association between adult height and breast cancer supports this suggestion.5, 6

Most of the research on diet and breast cancer has focused on the possible effect of fat intake, which remains hotly debated.7–9 The consumption of dairy products, which are important contributors to the fat intake in many Western countries, has also yielded conflicting findings regarding breast cancer incidence.10, 11 Generally, no association between milk consumption and breast cancer incidence has been found in prospective studies,12–14 although both positive15 and negative16 findings are reported. As for any relationship between childhood or adolescent milk consumption and subsequent breast cancer, only limited data are available. A few case-control studies have, however, indicated that there may be no17 or an inverse association.18, 19

Several components of milk (e.g., growth factors, fatty acids, calcium) have been hypothesized to play a role in the development of breast cancer.20–22 In Norway, there is a generally high consumption of milk among both children and adults (per capita supply in 1998 was 142 kg),23 but also considerable variation in consumption. Any elucidation of the association between present or past milk consumption and breast cancer would be of significance. We therefore examined whether milk consumption, both as a child and as an adult, was associated with breast cancer incidence in a population-based prospective study characterized by a high milk consumption. Furthermore, we examined the effect of a sustained high or low milk consumption. All known reproductive and hormonal risk factors were adjusted for in the analyses.

MATERIAL AND METHODS

  1. Top of page
  2. Abstract
  3. MATERIAL AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. REFERENCES

Study population

In 1991–92, a random, nationwide sample of 100,000 Norwegian women born 1943–57 was drawn from the National Central Person Register and invited to participate in the Norwegian Women and Cancer Study (NOWAC). A total of 61,000 women were randomly sampled in 1991, and an additional 39,000 (Norwegian citizens only) in 1992. The women received a mailed letter of invitation requesting informed consent and a self-instructive questionnaire. A written reminder was sent to non-responders about 6 weeks after the first invitation. Altogether 57,664 women answered the questionnaire. Sixty women answered “No” to participating in the record linkage and were excluded from the analyses, giving a crude response rate of 57.6%. Corrected for non-completion (death, severe mental handicap, unknown address), the response rate was 58.4%. Statistics Norway was responsible for sampling and for mailing of the questionnaires.

A comparison of certain characteristics (age, education, parity) among 15,000 women that were invited to participate in the study in 1992 and those who responded (n = 9,237) revealed no major selection biases. Compared with the total sample of eligible women, the responders were slightly younger. The distribution of women in age groups 35–39 years, 40–44 years and 45–49 years were 35.9%, 32.2% and 31.9%, respectively for responders, and 34.3%, 32.8% and 32.9%, respectively for the total eligible sample. Among the responders, 26.0% of the women had 13 years or more of education; the corresponding figure for the total eligible sample was 21.9%. As for parity, 32.4% of the responders had 3 or more children, whereas for the eligible sample the figure was 32.0%. The percentage of nulliparous women was 8.6 among responders and 10.4 and among the eligible sample.

The study was approved by the Regional Committee for Medical Research Ethics and the Norwegian Data Inspectorate.

Assessment of milk consumption

The dietary part of the questionnaire was designed in a food frequency manner, asking about average intake of 28 food items, including alcoholic beverages, during the last year. The question about milk consumption as a child was asked as ‘How much milk did you drink as a child every day?’ with fixed answering categories: ‘none,’ ‘1–3 glasses,’ ‘4–6 glasses’ and ‘7 glasses or more.’ Adult milk consumption was accessed as 3 separate questions according to the fat content of the milk: whole milk (3.9% fat), low fat milk (1.5% fat) and skimmed milk (0.1% fat). Nine different answering categories were given, ranging from ‘almost never’ to ‘6–10 glasses per day.’ Total adult milk consumption was calculated by summarizing the consumption of milk from all three questions. If all 3 questions were left blank, the questionnaire was excluded from the analyses of adult milk consumption. If 1 or 2 of the milk questions were left blank, zero consumption of that particular type of milk was assumed. Intake of fat from milk was calculated by multiplying the number of glasses of each milk type by the respective fat content, and thereafter summarized.

Reproducibility of the milk consumption questions

A total of 555 women included in the cohort were asked to fill in the same questionnaire twice, about 4 months apart, and 341 women agreed to do so. Weighted kappa for milk consumption as a child was 0.54. When excluding women with missing values the kappa estimate was 0.60. For adult milk consumption, weighted kappa for whole milk consumption was 0.43, for low fat milk 0.55, and for skimmed milk 0.50. The kappa estimates rose to 0.80, 0.73 and 0.76 respectively, when women with missing values were excluded. Most missing values on either the test or the retest corresponded to ‘almost never’ on the other test occasion.

Identification of breast cancer cases

Information from the questionnaire was linked to the Cancer Registry of Norway to identify incident breast cancer cases. The accuracy of the linkage was ensured by the unique 11-digit identification number that all Norwegian citizens have. In Norway, it is mandatory by law to report all incident cancer cases to the cancer registry, and the registry has an almost complete record of all cancer cases.24, 25 All but 8 cases were histologically verified. A corresponding linkage to records at Statistics Norway provided information on death and emigration.

Person years of follow-up were calculated as the time elapsed from date of entry into the cohort (defined as 3 months after mailing of the invitation letter) to the time of cancer (any type), to time of death or emigration, or to the end of follow-up (31.12.1997), whichever came first.

Of the 100,000 women initially invited to participate in NOWAC, 6,000 were given a questionnaire without dietary questions.26 The responders of this questionnaire (n = 3,694) were not included in the present analyses. Furthermore, the following exclusions were made: 6 women died before the start of the follow-up; 986 women had a prior cancer diagnosis and 4 women were excluded as lost to follow-up due to change of identification number. Women who emigrated were followed until emigration, if their emigration date was known, otherwise they were excluded (n = 118). Finally, we excluded 101 women who did not answer any of the milk questions. A total of 52,695 women were thus available for follow-up analyses.

The number of postmenopausal women in the cohort was low, and the present analyses were restricted to premenopausal women. Women who were premenopausal at baseline were treated as premenopausal until they reached the age of 50 during follow-up, at which time they were censored. The age of 50 as a dividing line for menopausal status was chosen based on data from an older sub-cohort of NOWAC (data not shown). The ultimate cohort consisted of 48,844 women.

Statistical analyses

Cox proportional hazards regression analyses were carried out to investigate the simultaneous effect of milk consumption and co-variates on breast cancer incidence rate. Incidence rate ratios (IRR) and 95% confidence intervals (CI) were calculated. In multivariate analyses we adjusted for age, maternal history of breast cancer, age at menarche, number of children, age at first birth, current use of oral contraceptives, body mass index [BMI = wt(kg)/ht(m2)], years of education, level of physical activity and alcohol consumption. Adjustment for total energy intake was not possible due to the restricted number of food items in the questionnaire.

Childhood milk consumption was categorized as in the questionnaire (i.e., ‘none,’ ‘1–3,’ ‘4–6’ and ‘7 or more’ glasses of milk per day), with ‘none’ as the reference group. Adult milk consumption at baseline was divided into the categories ‘none,’ ‘0.1–1.0,’ ‘1.1–3.0’ and ‘3.1 or more’ glasses of milk per day, with ‘none’ as the reference group. The combined effect of childhood and adult milk consumption was examined by constructing a 3-level variable: ‘low consumption,’ defined as no milk consumption on at least one of the occasions and not more than next-lowest consumption on the other occasion; ‘high consumption,’ defined as the highest milk consumption on at least 1 of the occasions and not less than the next-highest consumption on the other occasions; and ‘moderate consumption,’ defined as all other combinations. The low consumption group was used as the reference category.

The assumptions of proportional hazards for the exposures of interest were examined and were not found to be violated. All reported p-values are two-sided, and a significance criterion of p < 0.05 was used. The number of subjects included in the separate analyses varies somewhat due to item non-response. Statistical analyses were done by means of the SAS software package, version 6.12.

RESULTS

  1. Top of page
  2. Abstract
  3. MATERIAL AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. REFERENCES

A total of 317 incident cases of premenopausal breast cancer were diagnosed among the 48,844 women during follow-up. The number of person-years was 303,529 and mean follow-up time was 6.2 years (range 0.04–6.63). The mean age at diagnosis was 44.6 years (range 35.8–50.0).

Some baseline characteristics of the cohort are given in Table I. The mean age at entry was 40.7 years. Ten percent of the women did not have any children, whereas of those who had children the average number of children was 2.3. Some 7% of the women reported not drinking milk as a child. Most of the women reported a moderate milk intake as a child, whereas 2% reported drinking 7 glasses or more per day. As for adult milk consumption (i.e., at baseline), nearly 10% reported not drinking milk. Low fat milk was most frequently used, followed by skimmed milk and whole milk. The women reported on average 1.7 glasses of milk per day, and the average intake of fat from milk was 3.2 grams per day.

Table I. Selected Baseline Characteristics by Age Group in the NOWAC Study
Characteristics34–39 years (n = 21,028)40–44 years (n = 16,312)45–49 years (n = 11,504)Total(n = 48,844)1
  • 1

    Subgroups may not total 48,844 because of item non-response.

Age, years (mean)36.642.046.540.7
Maternal history of breast cancer (%)3.74.35.04.2
Age at menarche (mean)13.213.313.413.3
Nulliparity (%)12.18.67.49.8
Number of children, parous women only (mean)2.22.42.52.3
Age at first birth, parous women only (mean)24.123.823.923.9
Current use of oral contraceptives (%)9.53.22.06.1
Body mass index (mean)22.623.023.623.0
Years of education (mean)12.712.111.512.2
Physical activity score (min. 1–max. 10, mean)5.75.75.65.7
Alcohol intake, g/day (mean)2.82.93.12.9
Milk consumption as a child, glasses/day (%)
 None6.36.77.06.6
 1–361.260.461.260.9
 4–630.730.929.730.5
 ≥71.92.02.11.9
Milk consumption as an adult, glasses/day (%)
 None9.09.410.49.5
 >0.1–1.036.442.745.340.6
 1.1–3.042.038.736.239.5
 >3.012.69.28.110.4
Skimmed milk, g/day (mean)136128128131
Low fat milk, g/day (mean)161146139151
Whole milk, g/day (mean)37394239

Possible risk factors for breast cancer such as age at menarche, parity, height and BMI were examined in age-adjusted analyses. Age at menarche was not significantly related to premenopausal breast cancer risk. A 6% reduction in breast cancer risk was found for each child (p for trend = 0.22). As for height, a 3% increase in breast cancer risk was found for each additional cm of height, whereas a higher BMI was negatively associated with risk of breast cancer (data not shown).

Milk consumption as a child tended to be negatively associated with premenopausal breast cancer in age-adjusted analyses, although not significantly (Table II). Adjusting for maternal history of breast cancer, age at menarche, number of children, age at first birth, current use of oral contraceptives, BMI, years of education, level of physical activity, and alcohol intake, in addition to age, changed the incidence rate ratios only slightly (Table II). The effect of milk consumption as a child on premenopausal breast cancer risk was, however, modified by age at baseline (χ2 = 24.13, d.f. = 6, p < 0.001). For women in the youngest age group, there was a negative linear trend in the incidence rate ratios that was highly significant (p for trend < 0.001), whereas no significant trend was observed in the other age groups (data not shown). No significant interactions were found between milk consumption as a child and number of children, or between milk consumption as a child and BMI.

Table II. Incidence Rate Ratios (IRR) (95% Confidence Limits) of Breast Cancer According to Milk Consumption as a Child in the NOWAC Study
Milk consumption as a child (glasses/day)Age-adjusted model (n = 48,362)Multivariate model (n = 42,624)1
No. of casesIRR (95% CI)p for trendNo. of casesIRR (95% CI)p for trend
  • 1

    Adjusted for age, maternal history of breast cancer, age at menarche, number of children, age at first birth, current use of oral contraceptives, body mass index, years of education, physical activity, and alcohol consumption.

Did not drink milk281.0211.0
1–31900.74 (0.50–1.10)1610.79 (0.50–1.24)
4–6910.71 (0.46–1.08)800.76 (0.47–1.24)
7+50.61 (0.24–1.58)0.1840.64 (0.22–1.87)0.36

Adult milk consumption also tended to be negatively associated with incidence of premenopausal breast cancer after adjusting for possible confounders (p for trend = 0.12) (Table III). Women reporting drinking more than three glasses of milk per day had a 44% lower incidence rate of breast cancer (IRR = 0.56, 95% CI 0.31–1.01) than women not drinking milk at all. Subgroup analyses (age at baseline, number of children, BMI) did not add any further information on the relationship between milk consumption and breast cancer incident rate (data not shown). Using reference categories with larger number of cases did not alter the findings (data not shown).

Table III. Incidence Rate Ratios (IRR) (95% Confidence Limits) of Breast Cancer According to Milk Consumption as an Adult in the NOWAC Study
Milk consumption as an adult (glasses/day)Age-adjusted model (n = 48,229)Multivariate model (n = 42,604)1
No. of casesIRR (95% CI)p for trendNo. of casesIRR (95% CI)p for trend
  • 1

    Adjusted for age, maternal history of breast cancer, age at menarche, number of children, age at first birth, current use of oral contraceptives, body mass index, years of education, physical activity and alcohol consumption.

Do not drink milk331.0291.0
0.1–1.01310.91 (0.62–1.34)1100.85 (0.56–1.28)
1.1–3.01290.95 (0.65–1.39)1090.85 (0.56–1.28)
3.1+210.60 (0.35–1.04)0.19180.56 (0.31–1.01)0.12

Examining the relationship with incidence rate of premenopausal breast cancer separately for each milk type, and using non-milk drinkers as a reference group, revealed a tendency of a negative association for low fat milk (p for trend = 0.12), whereas no trend was seen for skimmed milk (p for trend = 0.88) and whole milk consumption (p for trend = 0.84) (data not shown). The number of cases drinking whole milk only was, however, low (n = 26). Nevertheless, when merging all women drinking whole milk only into 1 category, irrespective of amount consumed, and comparing them with non-milk drinkers, there was still no significant difference in incidence rate of breast cancer (IRR = 1.13, 95% CI 0.66–1.94) (data not shown in table).

Milk consumption was also expressed in quartiles of fat from milk per day. Using the lowest quartile as the reference category in a model including all the possible confounding variables, produced (non-significant) incidence rate ratios of 0.93, 0.78 and 0.79 for the 2nd, 3rd and 4th quartiles, respectively (data not shown in table).

When combining milk consumption as a child and as an adult we observed a clear negative trend in premenopausal breast cancer incidence rate with increasing milk consumption (p for trend = 0.03) (Table IV). Compared with women who reported no or low consumption of milk on both occasions, women with moderate milk consumption had a reduced incidence rate of breast cancer of about 25%, whereas women with a high milk consumption on both occasions had a reduced incidence rate of about 50%.

Table IV. Incidence Rate Ratios (IRR) (95% Confidence Limits) of Breast Cancer According to Milk Consumption as a Child and as an Adult in the NOWAC Study
Milk consumptionAge-adjusted model (n = 47,747)Multivariate model (n = 42,276)1
No. of casesIRR (95% CI)p for trendNo. of casesIRR (95% CI)p for trend
  • 1

    Adjusted for age, maternal history of breast cancer, age at menarche, number of children, age at first birth, current use of oral contraceptives, body mass index, years of education, physical activity and alcohol consumption.

Low421.0361.0
Moderate2540.81 (0.58–1.12)2160.76 (0.54–1.09)
High150.54 (0.30–0.97)0.04130.51 (0.27–0.96)0.03

Performing the analyses after exclusion of 104 women who were diagnosed with cancer (any type) during the first year of follow-up did not substantially change the results (data not shown).

DISCUSSION

  1. Top of page
  2. Abstract
  3. MATERIAL AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. REFERENCES

The present prospective study suggests an inverse dose-response association between milk consumption and premenopausal breast cancer incidence in a high consumption population. The clearest association with breast cancer incidence was found when combining milk consumption as a child and as an adult.

Lately there has been a growing interest in intrauterine and childhood nutrition as risk factors for subsequent diseases, including cancer.27–29 Adequate data on past diet is, however, very difficult to obtain. Based on wholesale statistics, we know that the per capita consumption of milk in Norway in the 1950s, when the women were growing up, was high: about 205 kilos, of which 195 kilos was whole milk, the rest being skimmed milk.23 Our questionnaire included only a single question on childhood milk consumption, and the answering categories were rather high. We do not know how well the question reveals real differences in past consumption. Although no significant association between childhood milk consumption and breast cancer incidence was found in our study, one may speculate on a negative association. Also, adding information on childhood milk consumption to the analyses of adult milk consumption strengthened the inverse association with breast cancer. A negative association with breast cancer risk was found in a Canadian case-control study applying childhood whole milk consumption categories ranging from ‘very rarely’ to ‘daily,’18 and in a US case-control study of adolescent intake of dairy fat.19 No association was, however, found in another US case-control study examining dairy product consumption during adolescence and risk of breast cancer.17

Several case-control studies concerning adult milk consumption and breast cancer have supported no association10, 30–33 or a positive association,11, 34–37 although inverse associations have also been reported.19, 38–40

Cohort studies, regarded as the epidemiological study design least prone to bias, have generally been more focused on consumption of dietary fat and animal products than on milk consumption per se. Overall, these studies have shown no association between milk consumption and breast cancer,12–14, 41 apart from a positive association between whole milk consumption and breast cancer incidence in a Norwegian analysis.15 A thorough examination of consumption of dairy products and breast cancer incidence has been done by Knekt et al.16 in a Finnish cohort. In accordance with our findings, their examination also revealed a reduced incidence rate of breast cancer with increasing milk consumption in a high consumption population. In the Finnish study, women in the highest tertile of milk consumption had a 50% reduced incidence rate compared with women in the lowest tertile.

The contradicting results may indicate that any association between milk consumption and breast cancer is not a strong one. Still, one has to remember the methodological weaknesses of dietary assessment methods. A variety of dietary methods have been applied, and we do not know how valid many of them are (including our own). The distribution of milk consumption varies greatly between the study samples, and the definition of reference group differs accordingly. Also, the possibility of sufficient adjustment for potential confounders varies between the studies.

Our study contains information on all known relevant reproductive and hormonal factors, as well as height, weight, years of education, level of physical activity and alcohol intake. Adjustment for these factors had only minor influence on the incidence rate ratios. Also, subgroup analyses indicated that the results could not be explained by residual confounding. On the other hand, our questionnaire only asked about consumption of a limited number of food items, and we chose not to include any additional dietary variables or energy intake in the multiple analyses. Including consumption of fruits and vegetables in our analyses did not change the results. This is in accordance with the findings of Knekt et al.16 who did not observe any notably change in the negative association between milk consumption and breast cancer after adjustment for selected food items, nutrients and energy.

The strengths of the NOWAC study are the population-based approach, the large size of the cohort, the prospective design, the almost complete follow-up with regard to incidence of cancer, death, and emigration, and the opportunity to adjust for all established risk factors for breast cancer. Due to the relatively young age of the women at entry and the short follow-up, however, the number of breast cancer cases in certain subgroups is small. This reduces the statistical power of the analyses and the feasibility of performing subgroup analyses.

No major selection biases were found in our cohort, although the responders differed from the eligible sample at some points (see methods). Acknowledging these differences, for bias to occur, the relationship between exposure and outcome needs to be different for those responding to the study and those who are not.42 The associations between breast cancer and known risk factors (age at menarche, parity, height, BMI) in our cohort were as shown by several others.2, 5

In 1997, milk and milk products contributed 24% of the dietary intake of fat in the Norwegian diet.23 Dairy products contributed 40% of the intake of saturated fatty acids; milk alone contributed 11%. The issue of dietary fat and risk of breast cancer has still not been settled despite numerous efforts to do so. In our study, expressing milk consumption in terms of grams of fat per day produced non-significant incidence rate ratios of about 0.80–0.90 in both the 2nd, 3rd and 4th quartiles of consumption and no trend in incidence rate ratios were seen either for skimmed or whole milk consumption.

In addition to saturated fat, milk is also an important contributor of calcium, and it contains significant amounts of protein and several vitamins and minerals. Whether an association with breast cancer is connected with one or more of these factors, or even with presently unknown components in milk, are questions that are not answerable by the present analyses. Calcium intake, however, has previously been inversely associated with cancer risk, especially of colon cancer,43 but also breast cancer.44 Furthermore, laboratory studies have suggested a possible protective role for calcium (together with vitamin D) in the development of breast cancer through its effect on the mammary gland.22 Another interesting biological mechanism by which milk intake can reduce breast cancer risk is the one hypothesized for conjugated linoleic acid (CLA). CLA is mainly derived from dairy products, and has been shown to block both local growth and systematic spread of human breast cancer in animal studies.21 Other researchers have hypothesized that dairy products may increase breast cancer risk through their content of oestrogen and growth factors.20

In summary, in this prospective population-based study we found a negative association between milk consumption and premenopausal breast cancer that could not be explained by reproductive or hormonal factors. The association was not dependent on the fat content of the milk, leaving other milk components or even unmeasured lifestyle factors related to milk consumption as possible explanatory variables.

Acknowledgements

  1. Top of page
  2. Abstract
  3. MATERIAL AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. REFERENCES

We would like to thank Mrs. M. Albertsen and Mrs. K. Jensen for the data entry work, Mr. M. Ustad for programming and Dr. I. Thune for valuable comments on the manuscript.

REFERENCES

  1. Top of page
  2. Abstract
  3. MATERIAL AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. Acknowledgements
  7. REFERENCES