Dietary patterns and the risk of breast cancer in Japanese women

Abstract

Components of the Japanese diet that might contribute to the relatively low breast cancer incidence in Japanese women have not been clarified in detail. To evaluate associations between broad dietary patterns and breast cancer risk in a Japanese population, the authors conducted a case–control study using data from the hospital‐based epidemiologic research program at Aichi Cancer Center (HERPACC). Factor analysis (principal components) was used to derive food patterns based on the 31 food variables and allowed designation of four major dietary patterns: prudent, fatty, Japanese and salty. In total, 1885 breast cancer cases were included and 22 333 female non‐cancer patients were recruited as the control group. Odds ratios (ORs) and 95% confidence intervals (95% CIs) were determined using multiple logistic regression analysis. After adjusting for potential confounders, there were no clear associations between the fatty, Japanese or salty dietary patterns and overall breast cancer risk. In contrast, an inverse association was evident for the prudent dietary. Women in the highest quartile of the prudent dietary pattern scores, had a 27% decreased risk of breast cancer compared with those in the lowest (95% CI: 0.63–0.84, p for trend < 0.0001). In addition, for women with a body mass index ≥25, the highest quartile of the fatty factor score was associated with a 58% increment in breast cancer risk compared to the lowest quartile, with a significant linear trend (P = 0.027). The authors found the prudent dietary pattern to be negatively associated with breast cancer risk. In addition, the fatty and Japanese patterns may increase breast cancer risk among obese women. (Cancer Sci 2007; 98: 1431–1438)

Although the incidence of female breast cancer in most Asian countries is much lower than that in Western countries,^{( 1 )} there has been a marked increase in recent years. The age‐adjusted incidence per 100 000 women (standardized on the world population) in Japan (27.9, Osaka Prefecture) is approximately one‐quarter of that in the USA (103.9, Los Angeles county). The age distribution of breast cancer in Japanese women is very different from that in Western countries,^{( 1 )} that is, the drop in the age‐specific incidence curve for breast cancer follows behind the average age of menopause, even though the age‐dependent elevation of risk in premenopausal women is somewhat similar.

Much of the international variation is due to differences in established reproductive risk factors, but diet might also contribute and provide a potentially modifiable target for prevention. A case–control study the authors’ conducted detected a statistically inverse association between the risk of breast cancer and intake of soybean products or isoflavones,^{( 2 )} and a cohort study based on public health center in Japan found frequent miso soup and isoflavone consumption to be associated with a reduced risk of breast cancer.^{( 3 )} The authors also reported a significant decrease in postmenopausal breast cancer risk was observed for increasing intake for fish.^{( 4 )} A Japan Collaborative Cohort (JACC) study has found that an intake of fish fat or long‐chain n‐3 fatty acids was associated with a decreased risk of breast cancer among a Japanese population.^{( 5 )} However, any components of the Japanese diet that might underlie the relatively low breast cancer incidence observed in Japan have not been clarified in detail.

A number of previous epidemiologic studies have indicated that individual foods or nutrients, such as vegetables, fruits, fish, and soybean products, are possibly associated with the risk of breast cancer, although no particular dietary factor has shown strong or consistent results. Although most recent debate on the relationship between diet and breast cancer risk has been concentrated on specific nutrients and/or food, overall dietary patterns may be of greatest importance for primary prevention recommendations. People eat meals consisting of a variety of foods with complex combinations of nutrients, so that broad dietary patterns reflect many simultaneous dietary exposures. Factor analysis is a statistical tool that is increasingly being used to determine such patterns, although the results obtained so far have been inconsistent regarding risk of breast cancer.^{( 6} , ⁷ , ⁸ , ^{9 )} This might be because all of the studies hitherto have been performed with Western populations at high risk. The same approach with Japanese women experiencing a low risk of breast cancer may provide additional insights.

In the present study, the authors therefore identified major dietary patterns among Japanese women and examined their association with breast cancer using data from the hospital‐based epidemiologic research program at Aichi Cancer Center (HERPACC).

Materials and Methods

Data collection.  Details of the study design and subject characteristics have been described elsewhere.^{( 10} , ¹¹ , ¹² , ¹³ , ^{14 )} In brief, the HERPACC study has been conducted since 1988, with a self‐administered questionnaire being completed by all first‐visit outpatients to the Aichi Cancer Center Hospital (ACCH). The questionnaires are then collected after checking for incomplete responses by trained interviewers and the data loaded into the computer system of the Aichi Cancer Center Research Institute for linkage with the hospital‐based cancer registry files. This study was approved by the Institutional Review Board.

Among all first‐visit outpatients between January 1988 and December 2000 (n = 91 870), the questionnaire was administered to 83 642 patients. The remaining 8228 patients were excluded because of lack of questionnaires, that is, absence of the interviewer, age exclusion (younger than 18 years old), or a person other than the patient attended the consultation. Of the 83 642 patients, 82 553 (98.7%) completed the questionnaire adequately. Among all first‐visit outpatients, 16 249 patients (male, 7381; female, 8868) were detected using the hospital‐based cancer registry. ACCH is a typical domestic hospital, located in the central part of Japan, and 97% of outpatients are residents of the Tokai district area; 35% of outpatients were living in Nagoya city and 45% were residents of Aichi Prefecture other than Nagoya city.

The questionnaire includes questions on occupation, medical history, height, weight, weight at approximately 20 years of age, family history (parents and siblings), smoking and drinking habits, dietary habits, sleeping habits, physical exercise and reproductive history. The details are taken prior to assessment of symptoms and all information is collected before clinical diagnoses are made. A family history of breast cancer was considered positive if a mother and/or a sister had been diagnosed as having breast cancer.

Cases and controls.  The present analysis was restricted to women aged 40–79 years who visited the hospital between May 1990 and December 2000. Of 8868 female cancer patients, 1885 of whom were first diagnosed as having breast cancer on the basis of histological examination within 6 months of their first visit, were recruited as the case group. A methodological study applying the same HERPACC data set earlier showed that the odds ratios (ORs) based on a large number of controls gives more statistical power and steadier estimates than the use of matched controls,^{( 15 )} therefore, the authors used all non‐cancer individuals as candidates for controls in the present study. As the controls, 22 333 female first‐visit outpatients who had never been diagnosed as having cancer were thus recruited.

Dietary assessment.  All subjects in the present study were asked for average frequency and dietary habits during the period of 1 year before onset of the present disease or before the interview. The questionnaire included 13 general dietary factors, for example, dietary habits relevant to cooked rice, raw vegetables, fruit, milk, miso (soy bean paste) soup, between‐meal snacks, Japanese green tea, coffee, type of breakfast (cooked rice, bread, others), preference for fatty food or salty food (like, dislike), volume of meals (large, small) and dietary restrictions (yes, no). Participants also provided information on food frequency with reference to average consumption of 17 food items: pickles (hakusaizuke), salted vegetables (asazuke), dried/salted fish, cooked/raw fish, soy been curd (tofu), green leafy vegetables, carrot, pumpkin, cabbage, lettuce, potato, egg, chicken, beef, pork, ham or sausage, instant noodles. For drinking status, participants were asked about the frequency of consumption and the amount of consumption per occasion for five alcoholic beverages, that is, sake (Japanese wine), shochu (Japanese distilled beverage), beer, whiskey, and others. The amount of consumption per day was used in the estimation of total ethanol intake from these alcoholic beverages using the unit of Japanese sake (‘go’). One ‘go’ contains 27 g of ethanol.

Statistical analysis.  Factor analysis is a technique used to reduce a number of variables into fewer independent factors. Factor analysis (principal components) was used to derive food patterns based on the 31 food variables in the data. The authors conducted the factor analysis using the data of 22 333 female first‐visit outpatients who had never been diagnosed as having cancer. Dietary patterns were generated by factor analysis using SAS PROC FACTOR statistical software (SAS Institute). To achieve simpler structure with greater interpretability, the factors were rotated by an orthogonal transformation (Varimax rotation function in SAS), which maintains the uncorrelated nature of the factors and tries to get the original variables to load high on one of the factors and low on the remainder. In determining the number of factors to retain, we considered the eigenvalues, the scree test, and interpretability. Post‐rotated factor loadings revealed that four factors described well the distinctive dietary patterns of the study population.

We thus designated the four dietary patterns as: (i) prudent (mainly consisting of vegetables and fruit, soybean curd, fish and milk); (ii) fatty (mainly consisting of meat and fatty foods); (iii) Japanese (cooked rice for breakfast and miso soup); or (iv) salty (pickles, dried or salted fishes and salty foods). A factor score for each dietary pattern was calculated by weighting consumption of each food item by the corresponding factor loading and summing the resulting values.^{( 16 )} This score ranks individuals in terms of how closely they conform to the dietary pattern.

Logistic regression analysis was used to obtain ORs and 95% confidence intervals (95% CIs) as estimates of relative risk. The P‐value for trend corresponded to the estimate of the slope derived from the logistic model in the case that the integers, 1 to n, were assigned to the ordered n levels of each factor. The LOGISTIC procedure provided by SAS was used to perform the calculations. Multivariate models were adjusted for age (as a continuous variable), year of visit, motives for consultation (self recommendation, family recommendation, referral from other clinics, secondary screening after primary screening, others), menopausal status (premenopause, postmenopause), parity (0, 1, 2, 3+), age at first full‐term pregnancy (<20, 20–24, 25–29, ≥30 years), age at menarche (≤13, 14, ≥15 ), smoking status (never, ever, ≤10 cigarettes/day, >10/day), drinking status (never, ever, <22 g alcohol/day, ≥22 g alcohol/day), family history of breast cancer (yes, no), exercise (none, occasional, ≥3–4 times/month, ≥2 times/week), current body mass index (BMI; as a continuous variable). BMI was calculated as weight in kilograms divided by the square of the height in meters.

Quartiles of factor scores for each dietary pattern among controls were used for cutoff values. Multiple logistic regression was performed to estimate the OR and 95% CI of breast cancer according to quartiles of scores for each dietary pattern, taking the lowest quartile group as the referent group.

From the cross analysis between selected lifestyle factors among non‐cancer patients, there was much difference between younger and older age groups. In Japan, the westernization of lifestyle has progressed rapidly since the end of World War II. This remarkable change may be associated with the dietary intake. Lifestyle factors of women who were born before World War II may be different from those of women who were born after World War II. To stratify women according to younger and older age groups, the cut‐off point of either ‘equal to or above’ or ‘below’ the age of 50 years was used. In additional analyses, data were stratified by dichotomized current BMI. Because a BMI ≥25 is defined as obese by the Japanese Society for the Study of Obesity,^{( 17 )} it was adopted as the cut‐off for the higher BMI group. Analysis by smoking status and drinking status was also conducted.

Results

Table 1 shows factor loadings, which are equivalent to simple correlations between the food items and the dietary patterns. A positive loading indicates that the food item is positively associated with the dietary pattern, and a negative loading indicates an inverse association with the dietary pattern. Pattern 1 was labeled prudent because it reflected the correlated intakes of foods commonly thought to be healthy, such as vegetables and fruit, soybean curd, fish, milk and lower intake of fatty and salty foods. Pattern 2 was labeled fatty because it reflected the correlated intakes of meat, ham, sausage, and other fatty foods. Pattern 3 was labeled Japanese because it reflected the correlated intake of a typical Japanese diet, such as cooked rice for breakfast and miso soup and low intake of coffee. Pattern 4, which we called salty, is characterized by a high intake of salty foods, such as pickles, salty/dried fish.

Table 1.

Factor‐loading matrix for the four major dietary patterns identified using factor analysis

Food items	Factor I	Factor II	Factor III	Factor IV
	Prudent	Fatty	Japanese	Salty
Carrot	66	16	5	–6
Green leafy vegetables	65	9	3	2
Potato	53	23	10	–4
Pumpkin	52	6	12	–7
Cabbage	48	24	–19	19
Soy bean curd (tofu)	46	6	21	8
Fruit	45	–9	–5	7
Raw vegetables	45	0	–30	25
Cooked/raw fish	30	5	15	10
Milk	30	–2	–2	–9
Dietary restrictions	20	–16	7	–7
Japanese green tea	14	0	10	11
Pork	14	67	–2	–3
Beef	8	59	–11	0
Ham or sausage	3	52	–18	10
Chicken	26	51	6	–4
Egg	24	34	1	3
Preference for fatty food	–22	32	–4	19
Instant noodle	–16	28	–7	12
Between‐meal snacks	4	16	2	4
Volume of meals	–5	14	7	11
Type of breakfast	4	–1	60	21
Cooked rice	–3	7	51	23
Miso (soy bean paste) soup	24	2	36	26
Drinking	–13	9	–17	7
Coffee	–9	18	–36	2
Lettuce	36	22	–39	24
Pickles (hakusaizuke)	–1	4	7	50
Salted vegetables (asazuke)	11	4	5	50
Preference for salty food	–24	11	–3	40
Dried/salty fish	8	10	15	35
Variance explained by each factor	3.0118	1.9858	1.3988	1.2420

Values are multiplied by 100 and rounded to the nearest integer.

Characteristics according to quartiles of dietary pattern scores in control group are presented in Table 2. Age was positively associated with the prudent and Japanese dietary patterns, and inversely with the salty pattern. Women with high prudent and Japanese pattern scores were less likely to be smokers and drinkers, with a lower prevalence of nulliparous women. Women with a high prudent pattern score were more physically active, but this did not extend to the Japanese pattern. Women who scored high on the fatty and salty pattern exercised less and were more likely to be smokers and drinkers than those who scored low on these patterns.

Table 2.

Characteristics according to quartiles of dietary pattern scores in control group

	Dietary pattern	Quartile of dietary pattern score
		1 (lowest)	2	3	4 (highest)
Age (mean ± SD, years)	Prudent	51.4 ± 8.8	52.0 ± 8.9	52.6 ± 8.9	54.5 ± 9.3
	Fatty	56.3 ± 9.3	53.8 ± 9.0	51.3 ± 8.5	49.2 ± 7.8
	Japanese	50.3 ± 8.1	51.7 ± 8.8	52.9 ± 9.1	55.7 ± 9.3
	Salty	53.1 ± 9.4	52.1 ± 8.9	52.3 ± 8.9	53.1 ± 9.0
Body mass index (mean ± SD, kg/m 2 )	Prudent	22.6 ± 3.0	22.4 ± 2.9	22.3 ± 2.8	22.2 ± 2.8
	Fatty	22.4 ± 3.0	22.3 ± 2.8	22.3 ± 2.8	22.5 ± 2.9
	Japanese	22.0 ± 2.7	22.3 ± 2.8	22.5 ± 2.9	22.6 ± 3.1
	Salty	21.9 ± 2.8	22.3 ± 2.8	22.4 ± 2.9	22.8 ± 3.0
Current smoker (%)	Prudent	18.2	10.5	7.0	4.5
	Fatty	9.8	8.8	9.9	11.7
	Japanese	15.3	11.4	8.7	4.8
	Salty	8.5	9.7	10.3	11.7
Current drinker (%)	Prudent	33.1	27.8	24.3	18.6
	Fatty	19.2	24.0	27.7	33.0
	Japanese	36.8	28.7	24.1	14.2
	Salty	21.1	25.6	27.4	29.8
Exercise ≥2 times/week (%)	Prudent	11.8	16.1	18.5	24.1
	Fatty	19.4	17.8	17.2	16.2
	Japanese	18.4	17.7	17.4	17.2
	Salty	20.0	18.5	17.2	14.8
Age at menarche (mean ± SD, years)	Prudent	14.0 ± 1.7	13.9 ± 1.7	13.9 ± 1.7	14.0 ± 1.8
	Fatty	14.4 ± 1.8	14.0 ± 1.7	13.8 ± 1.6	13.6 ± 1.6
	Japanese	13.7 ± 1.6	13.9 ± 1.6	14.0 ± 1.7	14.3 ± 1.8
	Salty	13.8 ± 1.7	13.8 ± 1.6	14.0 ± 1.7	14.2 ± 1.8
Nulliparous women (%)	Prudent	13.7	8.3	7.6	8.1
	Fatty	14.7	10.2	7.8	5.0
	Japanese	11.0	9.8	8.3	8.6
	Salty	12.4	10.0	8.3	7.0
Postmenopausal women (%)	Prudent	47.4	50.4	52.6	60.1
	Fatty	70.2	57.9	46.4	36.0
	Japanese	41.3	48.5	53.3	67.5
	Salty	54.6	49.7	51.3	55.0
Family history of breast cancer (%) †	Prudent	4.0	4.1	4.3	4.1
	Fatty	4.1	4.2	4.0	4.2
	Japanese	4.4	4.4	4.0	3.7
	Salty	4.3	4.0	3.6	4.5
Proportion of selected food items (%)
Carrot (daily)	Prudent	1.1	6.2	18.1	59.1
	Fatty	17.2	18.0	20.0	24.0
	Japanese	17.1	18.5	20.1	24.0
	Salty	25.3	19.5	17.1	17.5
Pork (≥3 times/week)	Prudent	10.5	14.7	17.7	22.4
	Fatty	0.6	3.3	10.9	48.4
	Japanese	14.8	16.0	17.5	15.9
	Salty	17.2	16.2	15.6	15.0
Salted vegetables (asazuke) (≥3 times/week)	Prudent	20.3	25.7	28.9	32.7
	Fatty	26.0	27.3	26.6	26.4
	Japanese	22.6	24.2	29.1	31.1
	Salty	4.2	13.0	29.4	58.7
Miso (soy bean paste) soup (daily)	Prudent	45.8	58.2	66.0	75.1
	Fatty	58.6	59.8	61.2	62.4
	Japanese	33.4	54.8	72.8	85.3
	Salty	42.8	55.3	65.3	77.9
Proportion of cooked rice intake for breakfast	Prudent	42.3	44.9	46.0	47.5
	Fatty	48.7	44.9	43.3	43.3
	Japanese	2.5	24.8	68.6	91.9
	Salty	30.6	39.2	47.6	62.2

^† The presence of either a mother or a sister/sisters with breast cancer was considered a positive. SD, standard deviation.

After adjusting for potential confounders, there were no clear associations between the fatty, Japanese or salty dietary patterns and overall breast cancer risk (Table 3). In contrast, an inverse association was evident for the prudent dietary pattern in the multivariate‐adjusted risk factor mode. The association was strongest among women 50 years of age or older. Older women in the highest quartile of the prudent dietary pattern had a 34% decreased risk of breast cancer as compared with those in the lowest (95% CI: 0.55–0.79, P for trend < 0.0001). Among younger women, aged 40–49 years, the corresponding estimate was a statistically significant 21% decreased risk (P for trend = 0.06). The other three dietary patterns were not associated with breast cancer risk among younger or older women, and results in the subgroup analyses were similar to the overall results for these patterns (Table 3).

Table 3.

Odds ratios (ORs) and 95% confidence intervals (95%CIs) of breast cancer according to quartiles of factor scores for each dietary pattern

Dietary pattern	Quartiles of factor score				P for trend
	1 (low)	2	3	4 (high)
All women
Prudent
Cases/controls	565/5584	479/5584	450/5582	391/5583
Multivariate OR (95% CI)	1.00 (ref)	0.89 (0.78–1.01)	0.85 (0.74–0.97)	0.73 (0.63–0.84)	<0.0001
Fatty
Cases/controls	477/5578	519/5591	443/5578	446/5586
Multivariate OR (95% CI)	1.00 (ref)	1.12 (0.98–1.28)	0.97 (0.84–1.11)	0.99 (0.85–1.14)	0.414
Japanese
Cases/controls	440/5582	507/5588	483/5580	455/5583
Multivariate OR (95% CI)	1.00 (ref)	1.16 (1.01–1.33)	1.10 (0.96–1.27)	1.04 (0.90–1.20)	0.823
Salty
Cases/controls	472/5583	502/5584	466/5581	445/5585
Multivariate OR (95% CI)	1.00 (ref)	1.08 (0.95–1.24)	1.04 (0.91–1.19)	1.04 (0.90–1.19)	0.761
Age 40–49 years
Prudent
Cases/controls	222/2480	195/2472	204/2474	166/2478
Multivariate OR (95% CI)	1.00 (ref)	0.92 (0.75–1.13)	0.95 (0.77–1.17)	0.79 (0.63–0.98)	0.060
Fatty
Cases/controls	211/2477	198/2474	175/2476	203/2477
Multivariate OR (95% CI)	1.00 (ref)	0.96 (0.78–1.18)	0.84 (0.68–1.04)	0.99 (0.80–1.22)	0.646
Japanese
Cases/controls	180/2477	223/2477	202/2475	182/2475
Multivariate OR (95% CI)	1.00 (ref)	1.31 (1.07–1.62)	1.20 (0.97–1.48)	1.10 (0.88–1.38)	0.594
Salty
Cases/controls	195/2475	219/2480	188/2476	185/2473
Multivariate OR (95% CI)	1.00 (ref)	1.18 (0.96–1.45)	1.07 (0.86–1.32)	1.09 (0.88–1.36)	0.627
Ages 50–79 years
Prudent
Cases/controls	369/3107	249/3109	254/3101	226/3112
Multivariate OR (95% CI)	1.00 (ref)	0.70 (0.59–0.84)	0.73 (0.62–0.87)	0.66 (0.55–0.79)	<0.0001
Fatty
Cases/controls	278/3107	284/3116	270/3097	266/3109
Multivariate OR (95% CI)	1.00 (ref)	1.05 (0.88–1.26)	0.99 (0.83–1.18)	0.99 (0.82–1.19)	0.726
Japanese
Cases/controls	263/3109	299/3107	260/3111	276/3102
Multivariate OR (95% CI)	1.00 (ref)	1.13 (0.94–1.34)	1.00 (0.83–1.20)	1.05 (0.88–1.27)	0.922
Salty
Cases/controls	270/3109	289/3107	287/3103	252/3110
Multivariate OR (95% CI)	1.00 (ref)	1.06 (0.89–1.26)	1.08 (0.90–1.29)	1.00 (0.83–1.20)	0.971

Adjusted for age, visit year, motivation, body mass index, menopausal status, parity, age at first full‐term pregnancy, age at menarche, smoking, drinking, family history of breast cancer and exercise. ref, referent group.

Results of multivariate OR considering two categories of BMI (<25 and ≥25), are shown in Table 4. For women with BMI <25, the highest quartile of the prudent factor score was associated with a 30% reduction in breast cancer risk compared to the lowest quartile, with a significant linear trend (P < 0.001), whereas the corresponding estimate for women with BMI ≥25 was a statistically non‐significant 23% decreased risk (P for trend = 0.138). For women with BMI ≥25, those with highest factor scores for fatty and Japanese dietary patterns had a significantly higher OR for breast cancer. The OR in the top quartile of the fatty factor scores was 1.58 (95% CI: 1.14–2.19, P for trend = 0.027) with reference to the lowest quartile. The corresponding estimate for the Japanese dietary pattern was 1.45 (95% CI: 1.05–1.99, P for trend = 0.031). The risk with the salty dietary pattern did not differ between leaner and more obese women.

Table 4.

Odds ratios (ORs) and 95% confidence intervals (95% CIs) of breast cancer according to quartiles of factor scores for each dietary pattern, stratified by BMI

Dietary patterns	Quartiles of factor score				P for trend
	1 (low)	2	3	4 (high)
BMI <25
Prudent
Cases/controls	463/4669	373/4674	344/4667	314/4668
Multivariate OR (95% CI)	1.00 (ref)	0.83 (0.72–0.96)	0.78 (0.67–0.90)	0.70 (0.60–0.82)	<0.0001
Fatty
Cases/controls	389/4672	407/4663	351/4675	347/4668
Multivariate OR (95% CI)	1.00 (ref)	1.06 (0.91–1.23)	0.91 (0.78–1.06)	0.90 (0.76–1.05)	0.060
Japanese
Cases/controls	348/4665	501/5531	307/3811	338/4671
Multivariate OR (95% CI)	1.00 (ref)	1.26 (1.08–1.46)	1.13 (0.97–1.32)	1.02 (0.86–1.20)	0.826
Salty
Cases/controls	376/4671	399/4669	374/4665	345/4673
Multivariate OR (95% CI)	1.00 (ref)	1.09 (0.94–1.26)	1.07 (0.92–1.25)	1.05 (0.89–1.22)	0.628
BMI ≥25
Prudent
Cases/controls	100/878	101/877	97/877	80/877
Multivariate OR (95% CI)	1.00 (ref)	1.02 (0.76–1.38)	1.00 (0.74–1.35)	0.77 (0.56–1.07)	0.138
Fatty
Cases/controls	87/877	108/877	83/880	100/875
Multivariate OR (95% CI)	1.00 (ref)	1.39 (1.02–1.89)	1.13 (0.82–1.57)	1.58 (1.14–2.19)	0.027
Japanese
Cases/controls	75/877	96/876	96/878	111/878
Multivariate OR (95% CI)	1.00 (ref)	1.26 (0.91–1.74)	1.28 (0.93–1.77)	1.45 (1.05–1.99)	0.031
Salty
Cases/controls	89/878	101/878	96/875	92/878
Multivariate OR (95% CI)	1.00 (ref)	1.22 (0.90–1.66)	1.21 (0.88–1.65)	1.19 (0.87–1.64)	0.308

Adjusted for age, visit year, motivation, menopausal status, parity, age at first full‐term pregnancy, age at menarche, smoking, drinking, family history of breast cancer and exercise. ref, referent group.

For the prudent pattern, the multivariate OR was 0.68 (95% CI: 0.58–0.79, P for trend < 0.001) comparing top to bottom quartiles among non‐smokers. Among current smokers, the corresponding estimate was a statistically non‐significant 19% decreased risk (95% CI: 0.53–1.23, P for trend = 0.313; data not shown). For the other three dietary patterns, there were no differences between results for subsets (non‐smoker, smoker) and the groups overall. Results stratified by drinking status did not differ from the main analysis (data not shown).

Discussion

Four major dietary patterns were discerned in the population of Japanese women and it was found that the prudent pattern was negatively associated with breast cancer risk, particularly among relatively lean women above 50 years of age. In addition, the fatty and Japanese dietary patterns may increase breast cancer risk among obese women.

Weaknesses of the present study were that the dietary questionnaire did not include portion size and has not been validated, so it is not possible to estimate the intake of nutrients. The food items in the questionnaire were selected out of major food groups. The authors have confirmed that these food items were a major contributor in each food group among Japanese in the Tokai area, from which subjects were recruited.^{( 18 )} A strength is that the number of cases was relatively large. The Japan Public Health Center‐based prospective study identified the healthy, traditional and Western dietary patterns; they reported that the traditional and the Western dietary patterns were positively associated with colon cancer risk in women.^{( 19 )} Another study of dietary pattern among Japanese populations also observed the Japanese dietary pattern; however, there was no reported association between Japanese dietary pattern and colorectal adenomas.^{( 20 )} The present study is the first to evaluate the association between dietary patterns and breast cancer risk using factor analysis in Japanese women.

A previously published study of an Italian population, focused on the relationship between diet and breast cancer using factor analysis,^{( 21 )} and showed a ‘salad vegetable pattern’, characterized by high consumption of raw vegetables, had a significant protective effect against breast cancer. Regarding epidemiological results, a meta‐analysis confirmed the protect effect of the intake of vegetables against breast cancer risk from 26 published studies;^{( 22 )} meanwhile pooled prospective studies failed to detect any protective effect.^{( 23 )} The authors’ previous studies using HERPACC data also generated clear evidence that the risk of breast cancer in Japanese women is markedly decreased by a high intake of fruit and green–yellow vegetables (green leafy vegetables, carrots, pumpkin).^{( 4} , ^{12 )} This effect may be related, in part, to dietary fiber and antioxidant vitamins because of their biochemical and physiological effects. Green vegetables are rich in folic acid and folate deficiency can reduce the availability of S‐adenosylmethionine for DNA methylation and thereby influence gene expression.

Fish consumption among Japanese is much higher than in Western countries. An international ecologic study revealed that Japanese consume 6.21% of total energy as fish, while the figure for US whites is only 0.74%.^{( 24 )} Evidence of the anticarcinogenic effects of fish n‐3 polyunsaturated fatty acids on breast cancer has been provided by animal studies,^{( 25} , ²⁶ , ^{27 )} but the results from epidemiological studies that have examined the association between fish consumption and breast cancer risk in humans are inconsistent. Several case–control studies,^{( 28} , ²⁹ , ^{30 )} including the authors’ previous study,^{( 4 )} have suggested the protective effects of fish intake against breast cancer. In a Norwegian cohort frequent consumption of poached fish was associated with a decreased risk of breast cancer.^{( 31 )} A cohort study showed an inverse association between intake of fish fat and long‐chain n‐3 fatty acids and breast cancer risk among Japanese women.^{( 5 )} Other cohort studies, however, have found either no clear association,^{( 32} , ³³ , ^{34 )} or actually an increased risk.^{( 35} , ^{36 )} Mechanisms might include inhibition of cyclooxygenase and p21 gene expression and up‐regulation of p53 gene expression.^{( 37} , ^{38 )} Furthermore, the beneficial effect of fish intake may better capture the synergistic interactions of food compounds in the prudent dietary pattern in the present study.

Japanese frequently consume various forms of soybean products, including tofu (soybean curd), so that the diet is likely to be much richer in isoflavones than in the Western world. Soybeans in fact constitute a unique dietary source of a group of phytochemicals and several natural anticarcinogens, such as protease inhibitors, phytates, phytosterols, saponins and lignans. Furthermore, soybeans are an excellent source of dietary fiber and micronutrients, particularly calcium. The results are in line with the inverse association between intake of soybean products and breast cancer risk suggested from ecologic/cross‐sectional studies,^{( 39 )} and also from analytical investigations. Thus case–control studies have found that soybean food intake was associated with a decreased risk of breast cancer among premenopausal Singaporean women,^{( 40 )} and both pre‐ and postmenopausal Asian–American women,^{( 41 )} although a Chinese case–control study failed to detect any protective effects of soybean food.^{( 42 )} Cohort studies among Japanese,^{( 3} , ^{43 )} Japanese–American,^{( 44 )} and Caucasian–American women,^{( 45 )} have provided some evidence that soybean products may reduce the risk of breast cancer. The authors have already reported a reduction in breast cancer risk with high soybean consumption among Japanese women.^{( 2 )} The prudent dietary pattern identified in the present study implies an adequate intake of vitamins, dietary fiber, and a variety of biologically active compounds, all of which may contribute to the protective effect against breast cancer.

In the authors’ previous study a risk reduction with high intake of milk among premenopausal and postmenopausal Japanese women was observed.^{( 4} , ^{12 )} Experiments have suggested favorable effects of lactic acid‐forming bacteria on the intestinal microflora and the immune system.^{( 46 )} Some components of milk, for example, calcium and vitamin D, might play roles as anticarcinogenic agents. Calcium has been proposed to reduce fat‐induced cell proliferation by maintaining intracellular calcium concentrations.^{( 47} , ^{48 )} Vitamin D modulates calcium metabolism and has calcium‐independent antiproliferative actions.^{( 48 )} The prudent dietary pattern is loaded by a high intake of vegetables, fruit, fish, soybean curd and milk. The combination of these compounds may reduce the breast cancer risk among Japanese women.

The fact that the protective effect of the prudent dietary pattern was more pronounced among women 50 years of age or older is in line with the greater intake of the component food items among older women.^{( 49 )} For example, women 50 years of age or older consume 324.0 g per day of vegetables on average, while women 30–49 years of age consume 256.3 g. Average intakes of fruits are 161.6 g and 87.5 g, respectively.

From subgroup analysis the protective effect of the prudent dietary pattern was more prominent among women with BMI <25. Many studies have examined the relationship between anthropometric indices and breast cancer, and negative correlations among premenopausal women and positive correlations among postmenopausal women have been the predominant finding.^{( 50} , ⁵¹ , ^{52 )} The authors also confirmed the association of early adult body size and subsequent weight gain with the development of breast cancer among Japanese women.^{( 13 )} Women with a higher factor score of the prudent dietary pattern who tried to put health messages into practice by the adoption of appropriate behaviors were likely to be conscious of a feeling of health and to control their weight. Therefore, the finding that protective effects were strongest among lean women is in line with expectation. In addition, obese women may over‐report their vegetable intake.^{( 53 )} An Italian study,^{( 21 )} and a Breast Cancer Detection Demonstration Project follow‐up cohort study in the USA,^{( 54 )} showed reduced risks associated with salad vegetable and traditional southern patterns were more pronounced among women with a BMI <25, but protective effects were no longer present in overweight women. A possible explanation of the lack of the protective effect among obese women could be that adipocyte production of estrogen in obese postmenopausal women may mask a diet–breast cancer association.^{( 55} , ^{56 )}

We found no overall association between our fatty and Japanese pattern and the risk of breast cancer, although a positive link was evident among overweight women. When obese women (BMI ≥25) were stratified by menopausal status, only among postmenopausal women were the fatty and Japanese patterns associated with an increased breast cancer risk (data not shown). The present results are in agreement with previous evidence,^{( 57 )} that obesity is an important risk factor among postmenopausal women. Because these dietary patterns are associated with high energy intake, these dietary patterns might accelerate the risk of breast cancer among obese women due to extra‐ovarian estrogen production.

There are certain potential limitations in the present study that should be considered. One methodological issue is possible bias due to use of hospital‐based non‐cancer patients as referents. It is possible that there are discrepancies in patient characteristics compared with the general population, but outpatients in Japan, in general, visit hospitals directly when they have symptoms and/or some anxiety about their health. This situation is very different from that in the USA, where people visit local general clinics first, and are then referred to hospitals that function as secondary and/or specific facilities for further medical treatment. At the ACCH, incident cancer cases were found to comprise only 16% of all new female outpatients. Among randomly sampled non‐cancer outpatients (n = 1000) in 1988–1989, only 34% were found to have specific diseases, the most common being benign tumors and/or non‐neoplastic polyps (13.1%), mastitis (7.5%), digestive disease (4.1%), or benign gynecologic disease (4.1%).^{( 58 )} The authors have compared lifestyle characteristics between outpatients in ACCH and 1231 persons from the general population randomly selected from the Nagoya electoral roll, and confirmed that they are not substantially different.^{( 59 )} In women aged 50–59 years, the percentages of daily drinker of miso soup were 59.7% and 60.5% among community residents and first‐visit outpatients of ACCH, respectively. The corresponding percentages of daily fruit intake were 46.0% and 52.8%, respectively. The percentages of everyday intake of green tea were not markedly different between community residents (87.3%) and first‐visit outpatients (82.9%). It was concluded that, with due consideration of age, sex, and season in the analysis, it is feasible to use non‐cancer outpatients as referents in epidemiological studies. The present study was free of response information bias to the questionnaire because all data were collected prior to diagnoses. Eligible referents were not matched, because the authors’ previous study showed that a large number based on outpatients gives a steadier estimate.^{( 15 )}

Furthermore, there are some limitations due to applying factor analysis. Factor analysis involves decisions that can be called subjective or arbitrary, decision that can have some impact on both the results and their interpretation.^{( 16} , ^{60 )} The selection and grouping of foods for analysis, the methods of rotation, and the manner in which the factors are ultimately labeled, depend heavily on the subjectivity of investigators. Statistical analyses have a key role in nutritional epidemiology to confirm associations between dietary factors and breast cancer risk. The associations between dietary factors and breast cancer have not been clarified in detail.^{( 61 )} The reasons include methodological limitations,^{( 62 )} therefore we need to continually develop useful and new techniques. Factor analysis, although not commonly applied hitherto, appears to be a promising approach for the investigation of breast cancer with reference to diet, and useful for the combination of information on many dietary variables. The present study, to the authors’ knowledge, is the largest case–control investigation carried out so far on the topic, and identified a dietary pattern that is negatively associated with breast cancer risk, namely the prudent dietary pattern.

In conclusion, the results indicate that a dietary pattern characterized by frequent consumption of vegetables, fruits, fish, soybean curd and low fat intake is associated with a reduced risk of breast cancer in Japanese women. The authors’ finding that fatty and Japanese type diets may elevate the risk of breast cancer among obese women suggested the possibility of prevention of breast cancer by low fat and calorie intake plus successful weight control.