Dietary patterns and subsequent colorectal cancer risk by subsite: A prospective cohort study

Abstract

In order to investigate the associations between dietary patterns and the risk of colorectal cancer by subsite in Japan, the baseline data from a population-based cohort study of 20,300 men and 21,812 women were analyzed. We conducted factor analysis and identified 3 major dietary patterns, “healthy,” “traditional” and “Western,” and calculated the factor scores of each pattern for individuals. During 10 years of follow-up, 370 colorectal cancer cases were identified. We found a positive association between the traditional pattern and colon cancer risk in women [rate ratio for highest quartile (RR) = 2.06; 95% CI = 1.10–3.84; p for trend = 0.11], but not in men. This positive association was slightly stronger for proximal colon cancer (RR = 2.07; 95% CI = 0.84–5.12) than for distal colon cancer (RR = 1.84; 95% CI = 0.75–4.50). After multivariate adjustment, the Western dietary pattern was also positively associated with colon cancer risk in females (RR = 2.21; 95% CI = 1.10–4.45), with the strongest associations being observed for females with distal colon cancer (RR = 3.48; 95% CI = 1.25–9.65). We did not observe any significant association between the healthy dietary pattern and colon cancer risk. For rectal cancer, no significant associations were found for the 3 dietary patterns. In conclusion, we found that the traditional and the Western dietary patterns were positively associated with colon cancer risk in females. © 2005 Wiley-Liss, Inc.

Initial evidence from migrant and ecologic studies suggested that variations in colorectal cancer (CRC) incidence rates over time and between highly Westernized and less Westernized countries may be explained largely by changes or differences in environment factors, mainly dietary factors.1, 2, 3, 4, 5 A number of previous epidemiologic evidences have indicated that individual foods or nutrients,6, 7, 8, 9, 10, 11, 12 such as vegetables, fruits, fiber, animal fats, red meats, calcium, vitamin D, folate and Japanese traditional salty foods,13 are possibly associated with the risk of colorectal cancer, but the results to date have not been completely consistent. Most previous epidemiologic studies have focused on individual nutrients and/or food. Only a few studies14, 15, 16 have evaluated the relation of overall dietary pattern to the risk of CRC. A dietary pattern approach may provide additional insights that take into account the combined effects of foods. People eat meals consisting of a variety of foods with complex combinations of nutrients, not isolated nutrients. Because of the complexity of diets, the traditional approach with a single nutrient may potentially be confounded by the interactions between food components that are likely to be interactive or synergistic.17 The overall dietary pattern that reflects many simultaneous dietary exposures may be an important complementary approach for elucidating relationships between diet and health.

Colorectal cancer incidence rates among Japanese were relatively lower than those among other developed countries, but the mortality and incidence rates of colorectal cancer have been gradually increasing in recent years.18 These chronologic variations of CRC incidence and mortality rates in Japan could be associated with the change in dietary habits such as Westernization of diet [increase of animal foods (meat) consumption and decrease of grains simultaneously].19

The difference in incidence of CRC by subsite and gender has been determined in previous studies,10, 20, 21, 22, 23 which indicate that there are different risk factors associated with proximal (right) and distal (left) colon carcinogenesis and with gender. Accordingly, using factor analysis in the present investigation, we identified the dietary patterns and evaluated the associations between dietary patterns and colon and rectal cancer risks in a population-based cohort study, the Japan Public Health Center (JPHC)-based prospective study on cancer and cardiovascular diseases (JPHC Study Cohort I).

Material and methods

Study cohort

The JPHC Study Cohort I is a population-based prospective study launched in 1990. The study cohort included 54,498 residents (27,063 men and 27,435 women) from 14 administrative districts supervised by 4 public health centers (PHCs): Ninohe PHC area of Iwate Prefecture, Yokote PHC area of Akita, Saku PHC areas of Nagano and Ishikawa PHC area of Okinawa. Study population was defined to be all inhabitants in the study areas aged 40–59 years at the beginning of the study (1 January 1990). The study design has been described in detail previously.24 The JPHC study was approved by the institutional review board at the National Cancer Center.

Baseline questionnaire

The self-administered food-frequency questionnaire (FFQ) includes 44 food groups that were commonly consumed in this study population. Participants indicated their average frequency of consumption for each food group over the past month. For rice, inquiry was made as to the number of bowls consumed per day. The frequency of miso (fermented soybean paste) soup consumption was classified into 4 categories: rarely (< 1 day/week), 1–2 days/week, 3–4 days/week and almost every day (6 or more days/week), and the number of bowls per day was asked in the same manner as for rice intake. The frequency of other food group items was classified into 4 categories: rarely (< 1 day/week), 1–2 days/week, 3–4 days/week and almost every day (6 or more days/week). For each of the 9 nonalcoholic beverage items (green tea, Chinese tea, black tea, other teas, coffee, milk, soda, fruit juice, vegetable juice), the intake frequency was asked using 6 categories: rarely (< 1 day/week), 1–2 days/week, 3–4 days/week, 1–2 cups/day, 3–4 cups/day and 5 or more cups/day. Questions on the consumption frequency of 5 alcoholic beverages (sake, shochu, beer, whiskey and other) covered 6 categories (almost never, 1–3 days per month, 1–2 days per week, 3–4 days per week, 5–6 days per week, almost every day). The selected frequency category for each item was converted to a weekly intake. In calculating the amount of each food item and nutrients, we used the serving size based on the observed median values from the 14–28 day diet record data.25 Nutrients values were adjusted for total energy intake using the residual method.26 The diet record data were also used to assess the validity of the questionnaire. The validity and reproducibility of the FFQ used in this study were reported previously.27

In addition, participants were asked to respond to a self-administered questionnaire on lifestyle such as sociodemographic characteristics, leisure-time physical activity, medical history, use of vitamin supplements, family history of diseases and their history of cigarette smoking and alcohol consumption. A self-administered questionnaire was distributed to 54,498 registered residents (27,063 men and 27,435 women) in 1990 and was collected from 20,665 (76%) men and 22,484 (82%) women. Of 43,149 subjects who responded to the questionnaire, subjects with a self-reported serious illness (cancer, ischemic heart disease, cerebrovascular disease, chronic liver disease) at baseline, and subjects who were not Japanese or had already moved away at baseline, were excluded in this study after confirmation during the follow-up period. Additionally, subjects who reported extreme total energy intake (upper 2.5% or lower 2.5%) and subjects who reported a past history of cancer (268 men and 598 women) were also excluded, leaving a total of 42,112 subjects (20,300 men and 21,812 women) eligible for the analysis.

Follow-up and identification of cancer cases

We followed all registered cohort subjects from 1 January 1990 to 31 December 1999. Incident cases of cancer occurring in the cohort have been identified through continuous surveillance of hospital records, population-based cancer registries and death certificates. This detailed follow-up procedure was described elsewhere.24 Cases of colorectal cancer were extracted from the JPHC cancer registry based on site codes [International Classification of Diseases for Oncology, 2nd edition (ICD-O-2) code: C180-C189 (colon) and C199, C209 (rectum)]. A total of 370 cases of colorectal cancer (231 males and 139 females) were documented with pathologically confirmed diagnoses such as adenocarcinoma, including adenocarcinoma in situ, occurring in 1990–1999 as of November 2000. For analyses of colon cancer by anatomic subsite, proximal colon cancers were defined as those occurring from the cecum up through the splenic flexure (n = 112). Distal colon cancers were defined as those occurring from the descending colon up through the sigmoid colon (n = 122). The location of colon cancers was not specified (n = 14).

Assessment of dietary patterns

Factor analysis (principal components) was conducted to derive dietary patterns based on the 44 food groups and beverages for men and women separately using the Factor procedure in SAS (version 8, SAS Institute, Cary, NC). The factors were rotated by an orthogonal transformation (Varimax rotation function in SAS) to achieve a simpler structure with greater interpretability. We considered components with an eigenvalue greater than 1.5, the Scree test and the interpretability of the factors. This served to limit the number of factors, as well as to identify more meaningful factors. After Varimax rotation, factor scores were saved from the principal component analysis for each individual. All data presented here are from the Varimax rotation. These scores were used for comparison with other lifestyle factors and to estimate associations with colorectal cancers. Factor scores were categorized into quartiles based on the distribution of study population for men and women separately. Retained dietary patterns were labeled on the basis of interpretation of the nutritional implications of the data and did not represent a priori intake patterns. When the whole cohort was randomly divided into 2 groups, the 3 major patterns were similar between the 2 groups and closely resembled those for the overall sample.

Statistical analysis

A Cox's proportional-hazards model was used to calculated the relative risks (RRs) for each quartile compared with the lowest quartile of each dietary pattern score using PROC PHREG of the SAS program (SAS Institute). In these analyses, age, body mass index, total energy intake, education level, physical activity and family histories of colorectal cancer were used as covariates. Smoking habit and alcohol consumption were further added in the multivariate models only for men. We tested for linear trends across categories of dietary patterns by assigning each participant the median value for the category and modeling this value as a continuous variable. All analyses were separately conducted for men and women. For colon cancer, subgroup analyses were performed for proximal and distal colon cancer. In a separate analysis, we also conducted analysis jointly classifying subjects by major dietary pattern and age (≥ 50), obesity status (body mass index ≥ 25) and smoking status. PHC areas were also added in the multivariate models only for the “healthy” and the “Western” dietary patterns, not for the “traditional” dietary pattern, because the associations with salted food intake were always attenuated after adjusting for PHC area, where intakes varied significantly between areas.

Results

The Scree plot of eigenvalues retained the 3 major patterns for men and women separately; thus, we identified the 3 dietary patterns in the final models. Factor-loading matrixes for the 3 major dietary patterns are listed in Table I. The larger the loading of a given food item to the factor, the greater the contribution of that food item to a specific factor, and a negative loading indicates negative association with the factor. Dietary pattern 1 was heavily loaded with vegetables, fruits, soy products, seaweeds, mushroom, milk, beans and yogurt and was called the healthy dietary pattern. Dietary pattern 2 was loaded with pickled vegetables, salted fish and roe, fish, rice and miso soup for both genders with a negative loading for bread and butter. Dietary pattern 2 was additionally loaded with alcoholic beverages (sake, shochu and beer) for men and was thus called the traditional dietary pattern. Dietary pattern 3 was loaded with meat, poultry, cheese, bread, butter and was called the Western dietary pattern. Although the order of their importance varied, and in some instances the load of specific food items and alcoholic beverages was not equal for men and women, the major dietary patterns identified separately for men and women proved to be rather similar.

Table I. Factor-Loading Matrix for the 3 Major Dietary Patterns Identified using Factor Analysis

	Male			Female
	Factor 1 (healthy)	Factor 2 (traditional)	Factor 3 (western)	Factor 1 (healthy)	Factor 2 (traditional)	Factor 3 (western)
Absolute values < 0.15 were not listed for simplicity.
Yellow vegetables	0.63			0.65
White vegetables	0.64			0.59
Green vegetables	0.58			0.54
Fruits	0.57			0.52
Seaweed	0.56			0.59
Potatoes	0.56			0.57
Yogurt	0.46			0.49
Mushroom	0.47			0.46
Soy and soy products	0.49			0.47
Milk	0.34			0.38
Eggs	0.38			0.36
Beans	0.29		0.31	0.31
Japanese tea
Salted roe		0.64			0.61	0.35
Pickled vegetables	0.30	0.57			0.65
Dried fishes	0.32	0.57			0.60
Salted gut		0.57			0.47	0.42
Miso soup		0.43			0.50
Rice		0.42			0.51
Fish and shellfish	0.36	0.43		0.31	0.48
Sake		0.56
Shochu		0.29
Beer		0.34	0.23
Dressing			0.26		−0.32	0.25
Bread		−0.48	0.25		−0.45	0.27
Butter		−0.40	0.40		−0.44	0.37
Mayonnaise	0.37		0.32	0.33		0.36
Cheese			0.48		−0.32	0.38
Beef			0.54			0.45
Pork			0.39			0.48
Poultry			0.40	0.23		0.45
Bacon			0.49			0.55
Liver			0.46			0.38
Soda beverages			0.35			0.42
Fruit juice			0.39			0.40
Vegetable juice			0.38			0.32
Instant noodles			0.34			0.31
Coffee			0.21		−0.31	0.26
Black tea			0.25			0.24
Noodles			0.24

The baseline characteristics of both men and women according to the quartile of dietary pattern scores are shown in Tables II and III, respectively. Among both men and women, participants with a higher healthy pattern score tended to have a higher educational level, to smoke less, to consume more vitamin A, carotenoids, vitamin C, fiber, fat, protein and to consume less alcohol. Participants with a higher traditional dietary pattern score were slightly older, likely to consume more sodium, have a family history of colorectal cancer and a lower educational level. The traditional dietary pattern was, especially among men, positively associated with a higher level of energy and rice intake, which is a staple food in Japan. Men with a high traditional dietary pattern score were more likely to smoke and drink alcohol. Participants with a higher Western pattern score were younger, likely to smoke and drink and more likely to have higher fat and vitamin A intakes.

Table II. Baseline Characteristics According to Quartiles of Dietary Pattern Score in Males

	Dietary pattern	Quartile of dietary pattern score
		1 (lowest)	2	3	4 (highest)
1 Mean ± SD. 2 Median. 3 Energy-adjusted nutrient intake by residual method.
Age (years)	Healthy	48.0 ± 5.81	49.2 ± 5.9	49.9 ± 5.9	50.7 ± 5.9
	Traditional	49.0 ± 5.9	49.1 ± 5.9	49.4 ± 6.0	50.3 ± 5.8
	Western	50.7 ± 5.8	49.5 ± 5.9	49.0 ± 6.0	48.7 ± 5.9
Body mass index (kg/m^2)	Healthy	24 ± 3	24 ± 3	23 ± 3	23 ± 3
	Traditional	24 ± 3	23 ± 3	23 ± 3	23 ± 3
	Western	23 ± 3	24 ± 3	24 ± 3	24 ± 3
Education (college or higher) (%)	Healthy	12.6	13.9	14.8	15.8
	Traditional	18.6	17.0	12.9	8.5
	Western	13.1	14.6	14.6	14.8
Leisure-time physical activity (≥ 1 time/week; %)	Healthy	15.2	16.1	18.0	20.8
	Traditional	21.2	19.4	16.1	13.3
	Western	13.6	16.9	18.3	21.3
Current smoker (%)	Healthy	60.8	54.9	51.8	45.1
	Traditional	44.4	50.9	57.2	60.2
	Western	48.9	53.9	53.7	56.3
Drinker (≥ 5 times/week; %)	Healthy	51.9	50.6	48.7	43.1
	Traditional	15.5	36.6	61.0	81.0
	Western	40.7	48.5	51.7	53.4
Family history of colorectal cancer (%)	Healthy	0.83	1.04	0.87	1.10
	Traditional	0.79	1.10	1.08	0.87
	Western	1.34	0.97	0.81	0.73
Total energy (kcal)2	Healthy	1,817	2,037	2,153	2,323
	Traditional	1,641	1,957	2,249	2,584
	Western	2,116	2,040	2,002	2,173
Nutrient intakes (energy-adjusted)23
Carbohydrates (g)	Healthy	316	318	316	312
	Traditional	325	322	313	302
	Western	337	320	308	294
Fiber (g)	Healthy	6.13	7.66	8.81	10.45
	Traditional	7.89	8.28	8.39	8.38
	Western	8.72	8.15	7.98	8.12
Protein (g)	Healthy	52.0	58.1	62.3	68.2
	Traditional	58.9	59.8	60.1	61.8
	Western	58.5	59.2	60.0	63.2
Fat (g)	Healthy	23.5	27.5	30.9	35.7
	Traditional	32.0	29.4	28.3	27.7
	Western	25.2	27.5	29.9	34.7
Alcohol (g)	Healthy	52.8	40.7	34.9	28.1
	Traditional	34.0	37.4	37.8	40.8
	Western	34.9	38.6	40.1	39.9
Sodium (mg)	Healthy	1,797	2,104	2,250	2,333
	Traditional	1,702	2,038	2,255	2,450
	Western	2,143	2,091	2,094	2,279
Vitamin A (IU)	Healthy	1,155	1,775	2,674	4,493
	Traditional	2,204	2,003	2,124	4,354
	Western	1,384	1,798	4,449	5,358
Carotenoid (mg)	Healthy	898	1,210	1,704	2,299
	Traditional	1,635	1,527	1,448	1,488
	Western	1,409	1,373	1,519	1,705
Vitamin C (mg)	Healthy	42.3	58.6	71.5	89.8
	Traditional	59.8	64.2	67.3	69.8
	Western	65.5	62.4	63.9	70.9

Table III. Baseline Characteristics According to Quartiles of Dietary Pattern Score in Females

	Dietary pattern	Quartile of dietary pattern score
		1 (lowest)	2	3	4 (highest)
1 Mean ± SD. 2 Median. 3 Energy-adjusted nutrient intake by residual method.
Age (years)	Healthy	48.9 ± 5.91	49.4 ± 5.8	49.7 ± 5.8	50.4 ± 5.9
	Traditional	48.6 ± 5.9	49.1 ± 5.9	49.8 ± 5.9	50.9 ± 5.5
	Western	50.9 ± 5.7	49.7 ± 5.8	49.1 ± 5.9	48.8 ± 5.8
Body mass index (kg/m^2)	Healthy	24 ± 3	24 ± 3	23 ± 3	23 ± 3
	Traditional	24 ± 4	23 ± 3	23 ± 3	24 ± 3
	Western	24 ± 3	24 ± 3	24 ± 3	24 ± 3
Education (college or higher) (%)	Healthy	9.8	10.7	12.1	15.6
	Traditional	16.5	14.0	10.4	7.3
	Western	11.9	11.8	12.3	12.3
Leisure-time physical activity (≥ 1 time/week; %)	Healthy	9.32	12.1	15.7	20.9
	Traditional	16.8	16.5	13.9	10.8
	Western	14.5	13.8	14.0	15.7
Current smoker (%)	Healthy	9.1	5.0	4.9	3.8
	Traditional	7.4	6.4	4.9	4.0
	Western	3.9	5.4	6.1	7.4
Drinker (≥ 5 times/week)	Healthy	5.2	3.4	3.4	3.3
	Traditional	2.6	4.4	4.1	4.2
	Western	2.9	3.3	4.1	5.1
Family history of colorectal cancer (%)	Healthy	0.73	0.86	1.03	1.14
	Traditional	0.57	1.32	1.03	0.84
	Western	0.79	1.14	0.94	0.90
Total energy (kcal)2	Healthy	1,175	1,309	1,394	1,525
	Traditional	1,208	1,337	1,405	1,518
	Western	1,256	1,309	1,367	1,544
Nutrient intakes (energy-adjusted)23
Carbohydrates (g)	Healthy	223	213	207	200
	Traditional	206	209	211	214
	Western	218	212	208	202
Fiber (g)	Healthy	7.4	9.2	10.4	11.9
	Traditional	8.8	9.8	10.2	10.3
	Western	10.3	9.8	9.6	9.8
Protein (g)	Healthy	46.7	52.0	55.3	59.3
	Traditional	50.8	52.7	54.2	56.7
	Western	51.5	52.4	53.6	56.8
Fat (g)	Healthy	25.8	30.8	33.8	38.4
	Traditional	34.1	32.9	32.0	30.7
	Western	27.8	30.7	33.0	37.5
Alcohol (g)	Healthy	10.8	6.4	5.1	4.0
	Traditional	8.6	6.1	4.9	5.7
	Western	6.3	6.7	6.6	5.7
Sodium (mg)	Healthy	2,002	2,195	2,279	2,279
	Traditional	1,623	2,051	2,370	2,650
	Western	2,179	2,153	2,160	2,379
Vitamin A (IU)	Healthy	1,605	2,250	2,868	3,523
	Traditional	3,502	2,824	2,625	2,432
	Western	2,225	2,399	3,179	6,680
Carotenoid (mg)	Healthy	1,497	2,352	2,752	4,239
	Traditional	2,537	2,592	2,573	2,519
	Western	2,597	2,492	2,498	2,648
Vitamin C (mg)	Healthy	74.2	99.3	117.5	138.5
	Traditional	91.9	109.3	114.9	118.5
	Western	108.3	104.4	106.3	117.8

The relative risks (RR) and 95% confidence intervals (CIs) of each quartile of dietary patterns are shown in Table IV. After adjustment for the potential confounders, a higher traditional dietary pattern score was significantly associated with increased risk of colon cancer in females, but not in males. The multivariate-adjusted RR for the highest quartile of the traditional dietary pattern score in female was 2.06 (95% CI = 1.10–3.84) comparing with the lowest, though the test for linear trend was not significant. There was no substantial difference in this positive association between proximal colon cancer (RR = 2.07; 95% CI = 0.84–5.12) and distal cancer (RR = 1.84; 95% CI = 0.75–4.50). No significant associations were observed between the traditional dietary pattern and rectal cancer in either gender, although a higher traditional dietary pattern score was suggested to decrease the risk of rectal cancer in male (RR = 0.62; 95% CI = 0.28–1.39).

Table IV. Multivariate Relative Risks of Colorectal Cancer with 95% Confidence Intervals According to Quartile of the 3 Major Dietary Patterns: JPHC Study 1990–1999

	Quartiles (male)				p for trend	Quartiles (female)				p for trend
	1 (low)	2	3	4 (high)		1 (low)	2	3	4 (high)
1 Multivariate adjustment included age, body mass index, study area (for the healthy and Western dietary pattern), energy intake, education level, physical activity, family history of colorectal cancer, smoking status (for males) and alcohol consumption (for males).
Healthy dietary pattern
Person-years	47,265	47,481	47,622	47,710		52,124	52,338	52,530	52,289
Colorectal cancer
Cases	60	57	61	53		37	37	29	36
Multivariate1	1.00	0.88 (0.60–1.30)	1.01 (0.69–1.48)	0.81 (0.52–1.24)	0.80	1.00	0.94 (0.58–1.51)	0.80 (0.48–1.32)	0.98 (0.58–1.65)	0.82
Colon cancer
Cases	37	39	44	36		24	23	25	20
Multivariate	1.00	0.97 (0.60–1.54)	1.13 (0.71–1.80)	0.83 (0.49–1.41)	0.62	1.00	0.88 (0.48–1.60)	1.02 (0.57–1.85)	0.76 (0.39–1.50)	0.68
Proximal colon
Cases	16	15	19	17		15	9	11	10
Multivariate	1.00	0.73 (0.35–1.52)	0.90 (0.45–1.82)	0.68 (0.32–1.47)	0.83	1.00	0.45 (0.18–1.13)	0.67 (0.29–1.52)	0.47 (0.18–1.23)	0.96
Distal colon
Cases	18	21	21	19		8	14	11	10
Multivariate	1.00	1.18 (0.61–2.29)	1.36 (0.69–2.68)	1.10 (0.52–2.36)	0.93	1.00	1.84 (0.76–4.45)	1.48 (0.57–3.82)	1.50 (0.53–4.21)	0.62
Rectal cancer
Cases	23	18	17	17		13	14	4	16
Multivariate	1.00	0.74 (0.38–1.43)	0.79 (0.40–1.54)	0.76 (0.37–1.58)	0.76	1.00	1.05 (0.48–2.30)	0.33 (0.11–1.05)	1.43 (0.62–3.28)	0.34
Traditional dietary pattern
Person-years	46,883	47,043	47,711	48,441		51,611	52,011	52,487	53,172
Colorectal cancer
Cases	48	51	59	73		31	32	29	47
Multivariate	1.00	0.89 (0.58–1.36)	0.91 (0.58–1.42)	0.88 (0.55–1.42)	0.70	1.00	1.06 (0.64–1.76)	0.96 (0.57–1.63)	1.53 (0.93–2.52)	0.23
Colon cancer
Cases	32	39	37	48		18	21	18	35
Multivariate	1.00	1.06 (0.63–1.76)	1.01 (0.58–1.76)	1.05 (0.58–1.90)	0.68	1.00	1.20 (0.63–2.29)	1.04 (0.53–2.05)	2.06 (1.10–3.84)	0.11
Proximal colon
Cases	12	18	15	22		8	10	7	20
Multivariate	1.00	1.17 (0.53–2.56)	0.91 (0.39–2.16)	1.06 (0.44–2.55)	0.97	1.00	1.17 (0.45–3.07)	0.75 (0.25–2.20)	2.07 (0.84–5.12)	0.20
Distal colon
Cases	17	17	21	24		10	11	9	13
Multivariate	1.00	0.91 (0.43–1.92)	1.24 (0.58–2.63)	1.21 (0.53–2.77)	0.26	1.00	1.25 (0.53–2.99)	1.12 (0.44–2.84)	1.84 (0.75–4.50)	0.53
Rectal cancer
Cases	16	12	22	25		13	11	11	12
Multivariate	1.00	0.59 (0.28–1.29)	0.73 (0.35–1.54)	0.62 (0.28–1.39)	0.87	1.00	0.87 (0.39–1.97)	0.84 (0.37–1.94)	0.85 (0.36–2.02)	0.84
Western dietary pattern
Person-years	47,649	47,862	47,309	47,258		52,618	52,302	52,274	52,086
Colorectal cancer
Cases	64	59	55	53		36	37	32	34
Multivariate	1.00	0.98 (0.67–1.43)	0.96 (0.65–1.41)	0.93 (0.62–1.41)	0.85	1.00	1.31 (0.81–2.12)	1.22 (0.73–2.03)	1.45 (0.85–2.48)	0.59
Colon cancer
Cases	39	43	39	35		19	27	22	24
Multivariate	1.00	1.15 (0.72–1.82)	1.11 (0.69–1.79)	1.05 (0.63–1.75)	0.73	1.00	2.01 (1.07–3.81)	1.80 (0.92–3.52)	2.21 (1.10–4.45)	0.74
Proximal colon
Cases	16	19	18	14		12	13	10	10
Multivariate	1.00	1.37 (0.67–2.78)	1.37 (0.66–2.85)	1.17 (0.53–2.56)	0.61	1.00	1.89 (0.78–4.61)	1.62 (0.62–4.20)	1.66 (0.60–4.64)	0.87
Distal colon
Cases	20	22	17	20		6	12	11	14
Multivariate	1.00	1.04 (0.55–1.98)	0.89 (0.44–1.77)	1.10 (0.55–2.20)	0.84	1.00	2.15 (0.80–5.77)	2.18 (0.79–6.03)	3.48 (1.25–9.65)	0.30
Rectal cancer
Cases	25	16	16	18		17	10	10	10
Multivariate	1.00	0.71 (0.37–1.37)	0.70 (0.36–1.39)	0.73 (0.36–1.46)	0.87	1.00	0.66 (0.30–1.46)	0.68 (0.30–1.55)	0.77 (0.32–1.83)	0.64

After multivariate adjustment, the Western dietary pattern was also positively associated with the risk of colon cancer but only in females. The multivariate-adjusted RR for the highest quartile of the Western dietary pattern score in female was 2.21 (95% CI = 1.10–4.45) comparing with the lowest, although the test for linear trend was not significant. When the upper 3 quartiles were combined and compared to the lowest quintile, the Western dietary pattern was associated with a significantly higher risk (RR = 1.98, 95% CI = 1.11–3.52 in colon cancer; RR = 2.43, 95% CI = 1.01–5.91 in distal colon cancer; results not shown). The positive association was slightly stronger for distal colon cancer (RR = 3.48; 95% CI = 1.25–9.65) than for proximal cancer (RR = 1.66; 95% CI = 0.60–4.64). Fiber intakes (total, water-soluble, and water-insoluble) have been reported to be inversely associated with the risk of colorectal cancer.28 To examine whether the positive association for the Western dietary pattern was mediated by these nutrients, we included fiber intakes in the multivariate model. Further adjustment for fiber intakes did not alter this positive association for the Western dietary pattern score of females substantially: the RR for the comparison of the highest with the lowest quartile was 3.60 (95% CI = 1.29–10.1) in distal colon. No significant association was observed between the Western dietary pattern and rectal cancer in either gender.

Evaluation of dietary patterns with regard to risk of colon and rectal cancers showed that no apparent associations were found between the healthy dietary pattern and colorectal cancer risk, although nonsignificant inverse associations were suggested to exist in proximal colon cancer. The multivariate-adjusted RRs for the highest versus the lowest quartile of the healthy dietary pattern score in proximal colon cancer were 0.47 (95% CI = 0.18–1.23) in females and 0.68 (95% CI = 0.32–1.47) in males.

In a subgroup analyses stratified by known risk factors, the associations between the 3 dietary patterns and risk of colon and rectal cancers were generally consistent across different strata according to smoking status, body mass index and age. However, the associations with the Western dietary pattern in distal colon cancer were slightly stronger among male smokers and among females below age 50 years. For distal colon cancer, the multivariate adjusted RRs for the highest quartile of the Western dietary pattern compared with the lowest were 2.58 (95% CI = 0.93–7.18) in male smokers and 4.17 (95% CI = 1.05–16.6) in younger females (results not shown).

Discussion

The relationship between dietary pattern and the risk of colon and rectal cancer was examined in a population-based prospective study of 42,112 Japanese in the JPHC study. We identified the 3 distinct dietary patterns, healthy, traditional and Western, in this population and found that the traditional and Western dietary patterns were positively associated with colon cancer risk in female, independent of other lifestyle variables, whereas the healthy dietary pattern was not associated with risk. The positive association between the traditional dietary pattern and colon cancer risk was more prominent among women with proximal colon cancer, while the positive association between the Western dietary pattern and colon cancer risk was more prominent among women with distal colon cancer.

For the last 5 decades, the incidence and mortality rates of colorectal cancer have been sharply increasing in Japan, especially those due to distal colon cancer in the urbanized areas rather than in rural areas, but those due to proximal colon cancer have been mostly stable.29, 30 Furthermore, proximal colon cancer in females was more prevalent in rural areas than in urban areas, where distal colon cancer was prominent.30 Secular changes and geographic variations of the incidence and mortality rates due to colon cancer by subsite may imply that the risk factors for right-sided (proximal) colon cancer may differ from those for left-sided (distal) colon cancer. It is noteworthy that there may be different risk factors in the etiology of right colon cancer in rural areas and left colon cancer in urbanized areas in Japan. The rapid rise in CRC incidence parallels the changes in eating habits and Westernization of lifestyle among Japanese.19 Dietary changes over the last several decades, especially reduced intake of carbohydrates and salted foods, increased intake of fats and animal foods, as well as Westernized dietary practice, were considered the major plausible explanations for increasing trends in colorectal cancer in Japan.18, 19 The present result also suggested that the recent Westernization of eating habits of Japanese might be associated with the recent increase in mortality rates for distal colon cancer in Japan. From the present findings, it was suggested that the association of the Western dietary pattern with distal colon cancer was more apparent especially among females below age 50 and may reflect the change to Westernized eating habits, despite the small number of cancers in this subgroup analysis. Thus, interpretations should be made cautiously. Furthermore, it is important to note that we cannot always explain the time trend of disease by a single factor, and the findings from ecologic and cohort studies are sometimes inconsistent. Also, the results should be verified in a larger prospective study.

Direct comparison between the present and previous findings from the individual nutrient/food approach is difficult. Because there are many potential differences in nutrients between dietary patterns, this approach cannot be specific about the particular nutrients responsible for the observed differences in disease risk and thus it may not be very informative about any biologic relationship between dietary components and disease risk. Nevertheless, our findings are consistent with the previous findings of associations of single nutrients and foods identified in earlier studies, in which higher consumption of red meat,6 especially heterocyclic amines (HCAs),9 fat,6, 11 carbohydrates23 and salty foods13 has been associated with an increased risk of colorectal cancer. Furthermore, the dietary pattern approach is basically population-based, and if some foods were added or excluded, dietary patterns may be slightly changed. Thus, direct comparison even between the findings with dietary pattern analysis is difficult. Therefore, the results from the dietary pattern analysis should be interpreted carefully. Nevertheless, the dietary patterns identified in the present study were similar to those from previous studies among Japanese and Western populations. Interestingly, these 2 patterns, i.e., healthy and Western, were qualitatively similar to those of the Western population. The healthy pattern in the present study was also similar to the “healthy,”31 “vegetable and fruit”32 and “prudent”14, 16, 33 patterns identified in other studies. The Western pattern in our study was similar to those labeled “Western,”34 “Western breakfast” and “meat”32 among the Japanese population and the “Western” pattern among the U.S.14, 16 and Swedish33 populations. However, the traditional pattern was, as expected, the distinctive dietary pattern for Japanese, which included traditional staple foods, rice and many kinds of salted foods, such as pickled vegetables and salted fish and roe, which may be an indirect marker of genotoxic carcinogen exposure. The traditional pattern in the present study was similar to the “rice/snack” pattern identified among Japanese.32 In our previous report,35 the healthy and the traditional dietary patterns were significantly associated with the risk of gastric cancer. In fact, a similarity in the geographic distribution of colorectal and gastric cancers was identified in Japan. The correlation coefficients of age-adjusted mortality rate of colorectal and gastric cancer within 47 prefectures of Japan in 1995 were 0.43 in men and 0.33 in women (unpublished data).

To the best of our knowledge, only 3 previous epidemiologic studies have examined the associations between dietary pattern and CRC risk. In a case-control study in the of United States14 and a prospective study of U.S. women,16 the risk of colon cancer was positively associated with the Western dietary pattern and negatively associated with a “prudent” pattern. Even though our study population and their dietary characteristics were different from the previous findings in the U.S. population, the positive association between the Western dietary pattern and colon cancer risk is consistent with what was observed in previous studies of the Western dietary pattern. However, in a prospective study of Swedish women,15 there was no clear association between a Western, healthy, or drinker dietary pattern and CRC risk. However, these 2 previous cohort studies were limited only to women subjects. Although the healthy dietary pattern has been hypothesized to be associated with lower risk of colorectal cancer through the dietary antioxidant effects of β-carotene, vitamin C and vitamin E,8 it was not significantly associated with the risk of CRC in the present study.

This observed result gives support to a unifying hypothesis that diet and associated factors, such as physical activity and body size, increase the risk of colorectal cancer via their effects on serum insulin concentrations and on the bioavailability of insulin-like growth factor-I.36, 37 As for the most important among the environmental factors influencing the CRC risk, diets that are high in meat, saturated fats, refined carbohydrates and processed foods and low in vegetables, fruits and fiber have nearly all been associated with an increased risk of CRC.9, 23, 37 Diets high in carbohydrate, especially the digestible nonfiber portion of carbohydrates, may lead to a chronic state of elevated insulin and stimulate growth of colorectal tumors,23 although an excessive carbohydrate intake might not be linked with hyperinsulinemia in a healthy young subject. In addition, high carbohydrate consumption is closely linked to salty food consumption in Japan, as shown in the present study. A high consumption of salted fish, shellfish and vegetables has been reported to be associated with the increased risk of CRC.13, 38, 39 The mechanism, however, is unclear, and there is no evidence that nitrosamines contained in salt-preserved foods are involved in the development of colon cancer in humans. Nevertheless, N-nitrosamine has shown mutagenicity and carcinogenicity in laboratory animals.38

In the present study, we could not obtain consistent findings for dietary risk factors between male and female, although they were claimed to be risk or protective factors in Western populations. Given the male-female differences in incidence by subsite,22 it is likely that one or more risk factors might also exhibit differences such as those observed in this study. It was reported in a previous study40 that response to the dietary questionnaire might be more precise and reliable in females than in males, because females in Japan generally prepare the meals and are more conscious of diet and foods. However, in our previous study, there was no gender difference in the validity of a self-administered food frequency questionnaire in comparison with 28- or 14-day dietary records.25 The differential associations with the dietary pattern in men and women may be partially explained by the fact that other lifestyle factors, such as percentage of smokers and habitual drinkers, were substantially different within each dietary pattern between genders. The role of cigarette smoking and alcohol drinking may be strongly associated with the risk among men.41

The observed site-specific differences in risk between the genders, however, suggest possible differences in etiology for proximal and distal colon cancers that are consistent with women's higher incidence of proximal colon tumors and adenomas.22, 42 Distinct epidemiologic and clinicopathologic characteristics of CRCs based on their anatomical location suggest different risk factors and pathways of transformation associated with proximal and distal colon carcinogenesis. These differences may reflect distinct biologic characteristics of proximal and distal colonic mucosa, acquired in embryonic or postnatal development, that determine a differential response to uniformly distributed environmental factors.20

It is important to note that dietary patterns are associated with health behaviors, lifestyle and sociodemographic factors.43, 44 As shown in Tables II and III, major dietary patterns were associated with demographic factors and lifestyle habits. Therefore, we cannot exclude the possibility that not only dietary factors defined as dietary patterns, but also their related demographic and lifestyle factors, may affect the CRC risk, even though the related lifestyle variables were considered as potential confounders in the multivariate model.

This cohort study has been conducted in a large sample of men and women from the general Japanese population. One of its strengths is the high rate of participation and the completeness of follow-up, indicating that selection bias due to loss of follow-up is highly unlikely. Another strength of the prospective design is that the diet was measured before the disease was diagnosed, which diminishes the probability of recall bias of dietary intake. We have also controlled extensively for potential confounders. However, it is likely that unmeasured or unidentified risk factors may have affected the study results.

The present study had several limitations. Although our questionnaire requested detailed information regarding consumption of food and food groups, it was a short version that included only 44 food items. The number of total cohort subjects was not small, yet there were few cancer cases, particularly in the subgroup analysis. Accordingly, the risk estimates may be moderately imprecise, and more attention is needed to interpret the findings of the subgroup analysis.

In conclusion, the major dietary patterns of Japanese were identified using factor analysis, and the present findings indicated that the traditional and the Western dietary pattern were positively associated with colon cancer risk in females.

Acknowledgements

Mi Kyung Kim was awarded a Visiting Scientist Fellowship from the Foundation for Promotion of Cancer Research in Japan.

Appendix

The investigators and participating institutions in the JPHC Study Cohort I Group, a part of the JPHC Study Group (principal investigator: S. Tsugane), were as follows: S. Tsugane, S. Sasaki, Epidemiology and Biostatistics Division, National Cancer Center Research Institute East, Kashiwa; J. Ogata, S. Baba, National Center for Circulatory Disease, Suita; K. Miyakawa, F. Saito, A. Koizumi, Iwate Prefectural Ninohe Public Health Center, Ninohe; Y. Miyajima, N. Suzuki, S. Nagasawa, Akita Prefectural Yokote Public Health Center, Yokote; H. Sanada, Y. Hatayama, F. Kobayashi, H. Uchino, Y. Shirai, T. Kondo, Nagano Prefectural Saku Public Health Center, Saku; Y. Kishimoto, E. Takara, M. Kinjo, T. Fukuyama, Okinawa Prefectural Ishikawa Public Health Center, Ishikawa; S. Matsushima, S. Natsukawa, Saku General Hospital, Usuda; S. Watanabe, M. Akabane, Tokyo University of Agriculture, Tokyo; M. Konishi, Ehime University, Matsuyama; S. Tominaga, Aichi Cancer Center Research Institute, Nagoya; M. Iida, S. Sato, Center for Adult Diseases, Osaka; the late M. Yamaguchi and Y. Matsumura, National Institute of Health and Nutrition, Tokyo; Y. Tsubono, Tohoku University, Sendai; H. Iso, Tsukuba University, Tsukuba; H. Sugimura, Hamamatsu University, Hamamatsu; and M. Kabuto, National Institute for Environmental Studies, Tsukuba.