Dietary Intake of Fish, ω-3 and ω-6 Fatty Acids and Risk of Colorectal Cancer: A Prospective Study in U.S. Men and Women

Abstract

The association between fish, ω-3 and ω-6 polyunsaturated fatty acid (PUFA) intake and risk of colorectal cancer (CRC) remains inconclusive. Recent prospective studies suggest that the relationship may vary by gender, subsite and duration of follow-up. We followed 123,529 US adults (76,386 women and 47,143 men) without a history of cancer at baseline for 24 to 26 years. Fish and PUFA intake was assessed at baseline and updated every 4 years by using a validated food-frequency questionnaire. We found no overall association between fish, ω-3 and ω-6 PUFA intake and CRC risk with hazard ratio (HR) of 1.03 (95% confidence interval (CI): 0.89-1.20) comparing marine ω-3 intake of ≥0.30 g/d versus <0.15 g/d among women and 1.05 (95% CI: 0.85-1.30) comparing intake of ≥0.41 g/d versus <0.16 g/d among men. However, fish and marine ω-3 PUFA intake appeared to be positively associated with risk of distal colon cancer in both men and women, and inversely with risk of rectal cancer in men. In an analysis based on a limited number of cases, marine ω-3 PUFA intake assessed 12-16 years before diagnosis tended to be inversely associated with CRC risk in men (HR: 0.76; 95% CI: 0.52-1.10). In conclusion, although no overall association between fish, ω-3 or ω-6 PUFA intake was observed with CRC risk, marine ω-3 PUFA may be differentially associated with risk of distal colon and rectal cancers, and a long latency may be needed for its protection against CRC in men.

Introduction

Colorectal cancer (CRC) is the third most common cancer and the fourth leading cause of cancer death in the world¹. Experimental evidence supports the hypothesis that ω-3 polyunsaturated fatty acids [PUFAs, i.e., α-linolenic acid (ALA), eicosapentaenoic acid (EPA), docosahexaenoic acid (DHA), and docosapentaenoic acid (DPA)] protect against CRC, while ω-6 PUFAs [i.e., linoleic acid (LA) and arachidonic acid (AA)] promote CRC development^{2, 3}. The proposed mechanisms include modulation of inflammation, influence on cellular oxidative stress, cellular signaling, alteration of membrane dynamics and cell surface receptor function, and impact on insulin sensitivity^{2, 3}.

However, epidemiologic data are inconclusive. In some prospective studies, fish intake, an important source of ω-3 PUFAs, was inversely associated with risk of CRC^4-6, while in others a generally null association^7-12 or even positive relationship for intake of smoked and salted fish was reported ¹³. Recently, a meta-analysis of seven prospective studies showed insufficient evidence of a protective effect of ω-3 PUFA intake on overall CRC risk, although a potential benefit for men was observed¹⁴. Two prospective studies on circulating PUFAs also reported an inverse association between ω-3 PUFA levels and CRC risk among men^{15, 16}. Besides sex, duration of follow-up may also modify the association. Some studies have found a stronger association between baseline ω-3 PUFAs and reduced CRC risk in analyses with longer duration of follow-up, suggesting an early-acting role of ω-3 PUFAs against colorectal carcinogenesis^{17, 18}. Furthermore, accumulating evidence suggests risk factors of CRC vary by subsite¹⁹.

The only prospective study to date that separately examined proximal, distal colon and rectal cancer risk in relation to PUFA intake reported a stronger inverse association of marine ω-3 PUFAs with CRC in the proximal regions than in the distal regions²⁰. Finally, the overall amount of ω-3 PUFA intake may also matter to the ultimate effect. In a Japanese study, a U-shaped association was observed between marine ω-3 PUFA and risk of rectal cancer in men²⁰. Given the limited evidence, further studies are warranted.

Thus, we performed a detailed analysis on the association between fish, PUFA and CRC risk in two large prospective cohorts, the Nurses’ Health Study (NHS) and the Health Professionals Follow-up Study (HPFS). The prospective, repeated measurements of diet over 24-26 years as well as the large sample size provided us a unique opportunity to examine the association between these dietary factors and risk of CRC according to sex, timing of intake, anatomic site of the tumor, and level of intake.

Methods and Materials

Study population

The NHS and HPFS cohorts have been described in detail elsewhere^{21, 22}. In brief, the NHS included 121,701 registered female nurses who were aged 30-55 years at baseline in 1976 in the United States. The HPFS included 51,529 US male professionals who were aged 40-75 years at baseline in 1986. In both cohorts, questionnaires were administered biennially and follow-up rates have been approximately 90% in each 2-year cycle.

In the present study, we used 1984 for the NHS and 1986 for the HPFS as baseline, when dietary intake was assessed using a validated 116- to 131-item food frequency questionnaire (FFQ). Among participants who returned the baseline FFQs, we excluded those who had a history of cancer (except non-melanoma skin cancer) or ulcerative colitis, left more than 70 items blank on the FFQs, had missing information on fish, ω-3 or ω-6 PUFAs intake, or reported implausible energy intake levels (<600 or >3500 kcal/d for women, <800 or >4200 kcal/d for men). After exclusions, a total of 76,386 women and 47,143 men were included in the analysis. The study protocol was approved by the institutional review boards of Brigham and Women’s Hospital and Harvard School of Public Health.

Dietary assessment

Since baseline, we have updated dietary information from participants through repeated administration of extended versions of FFQs in 1986 and every 4 years thereafter in both the NHS and HPFS. In each FFQ, participants were asked how often, on average, they consumed each food of a standardized portion size during the previous year. Nine response options were provided, ranging from “never or less than once per month” to “6 or more times per day”. Questionnaire items about fish consumption included (a) dark meat fish (3-5 oz); (b) canned tuna (3-4 oz); (c) other fish (3-5 oz); and (d) shrimp, lobster, or scallops as main dish (3.5 oz). Since 1994 in the NHS and 1998 in the HPFS, consumption of breaded fish cakes, pieces, or fish sticks was additionally queried. The questionnaire also included a write-in section for food not listed and for the exact brand of margarine and type of fat used for frying, cooking, baking, and at the table. The average daily intake of nutrients was calculated by multiplying the reported frequency of consumption of each item by its nutrient content and then summing across from all foods. The nutrient composition data were based primarily on the US Department of Agriculture Nutrient Database that corresponded to each time when FFQs were administered, and supplemented by other published sources and personal communications from other laboratories and manufacturers. The calculation of ω-3 PUFA intake has been described previously in detail²³. Total ω-3 PUFA intake is the sum of intake of ALA, EPA, DHA, and DPA. Marine ω-3 PUFAs include EPA, DHA and DPA. Total ω-6 PUFAs include LA and AA. We adjusted nutrient consumption for total energy intake using the nutrient residual method²⁴.

The reproducibility and validity of FFQs in evaluating fish and PUFA intake have been assessed in a random sample of 118 Boston-area HPFS participants aged 45-70 years who completed two consecutive FFQs (in 1986 and 1987), two 1-week dietary records approximately 7 months apart, and provided subcutaneous fat aspirate samples²⁵. Spearman correlation coefficient estimated from the two FFQs was 0.61 for fish intake ²⁶ and 0.38 for energy-adjusted PUFA intake²⁷. The proportions of total fat represented by PUFA intakes measured by various approaches were almost identical (22.7% for questionnaire 1, 22.1% for questionnaire 2, 23.1% for dietary records, and 23.0% for fat aspirate)²⁵. The correlation of crude PUFA intake from the first and second FFQs with that from dietary records was 0.29 and 0.33, respectively. The corresponding values for energy-adjusted PUFA intake were 0.38 and 0.37²⁷. The energy-adjusted intake of EPA measured by FFQs correlated well with percentage of EPA in adipose tissue (Spearman correlation coefficient = 0.47; P < 0.001)²⁵. Similar findings were seen in an earlier validation study among women in the NHS cohort^{28, 29}.

Ascertainment of CRC cases

In both cohorts, self-reported diagnoses of CRC were obtained in biennial questionnaires, and participants who reported a diagnosis of CRC were asked for permission to acquire their medical records and pathologic reports. We identified deaths with over 96% sensitivity through the National Death Index and next-of-kin. For all colorectal cancer deaths, we requested permission from next-of-kin to review medical records. A study physician, blinded to exposure information, reviewed records to confirm CRC diagnosis and to extract information on anatomic location, stage, and histologic type of cancer. We documented a total of 1,469 incident CRC cases (713 proximal colon tumors, 416 distal colon tumors 310 rectal tumors, and 30 cases with unspecified location) in the NHS and 987 cases (342 proximal colon tumors, 302 distal colon tumors, 215 rectal tumors, and 128 cases with unspecified location) in the HPFS.

Measurement of erythrocyte PUFAs

Although the FFQ captures substantial information on food intake and has been validated for fatty acids using both dietary records and biomarkers, some PUFAs can also be derived from endogenous conversion of other PUFAs and a complementary assessment of biomarkers could aid in addressing potential measurement error. Thus we conducted a secondary analysis on the basis of predicted erythrocyte PUFA levels using previously measured erythrocyte PUFA among control participants in nested case-control studies of cardiovascular diseases and stroke in our cohorts. A prediction model for marine ω-3 PUFAs was developed by relating dietary intake to erythrocyte measurements. The details on blood collection and PUFA measurement have been described in previous publications^{30, 31}. In brief, we obtained blood specimen on ice packs by overnight courier from 32,826 women in the NHS between 1989 and 1990, and from 18,225 men in the HPFS between 1993 and 1995. Among participants who provided a blood sample, we used risk set sampling matched on age (within 2 years), time of blood donation and other factors (e.g., smoking and fasting status) to randomly select up to 2 controls for each new case with confirmed diagnosis of cardiovascular diseases and stroke during the follow-up from 1990 to 2006 in the NHS, and from 1994 to 2004 in the HPFS. After excluding outliers by the extreme Studentized deviate Many-Outlier procedure³², we included a total of 921 and 824 control participants for the present analysis in the NHS and HPFS, respectively. Fatty acid concentrations in erythrocytes were measured by gas-liquid chromatography at the laboratory of Dr. Hannia Campos at Harvard School of Public Health.

Assessment of covariates

In the biennial follow-up questionnaires, we collected and updated medical, lifestyle, and other health-related information, such as family history of CRC, body weight, cigarette smoking, physical activity, gastrointestinal endoscopic examination, and use of multivitamins, statins, aspirin and non-steroidal anti-inflammatory drugs (NSAIDs). Physical activity was assessed by summing the product of time spent on a variety of recreational or leisure-time activities with the average metabolic equivalent (MET) for that activity Menopausal status and postmenopausal hormone use were also ascertained in the NHS participants.

Statistical analysis

We calculated person-time of follow-up for each participant from the age in months at the date of the baseline questionnaire until the age in months at the date of death, CRC diagnosis, loss to follow-up, or end of follow-up (June 1, 2010 for the NHS, January 31, 2010 for the HPFS), whichever came first. We used time-varying Cox proportional hazards regression models to assess the associations of fish and PUFA intake with overall and subsite-specific risk of CRC within each cohort. To control as finely as possible for confounding by age, calendar time and any possible two-way interactions between these two time scales, we stratified the analysis jointly by age in months at the start of follow-up and calendar year of the current questionnaire cycle. To examine the differential associations with colorectal cancer by tumor subsite, we fitted a Cox proportional cause-specific hazards regression model using a duplication method^{33, 34}. P_{heterogeneity} by subsite was calculated using a likelihood ratio test by comparing the model in which the association with exposures was allowed to vary by subsite to a model in which all the associations were held constant. We used SAS 9.3 for all analyses (SAS Institute Inc., Cary, NC, USA).

In multivariate analysis, we simultaneously adjusted for several risk factors for CRC that may confound the relation of primary interest. These factors included family history of CRC, history of previous endoscopy, pack-years of smoking before age 30, body mass index(BMI, calculated as weight in kilograms divided by height in meters squared), physical activity levels, multivitamin use, regular aspirin/NSAID use, alcohol consumption, total calorie intake, and consumption of fiber, folate, calcium, vitamin D, red meat and processed meat. In the NHS, postmenopausal status and hormone use were additionally adjusted. We also conducted mutual adjustment to identify the independent associations between intake of ω-3 and ω-6 PUFAs and CRC risk. To represent long-term dietary and lifestyle patterns for a better control of confounding, we used cumulative averages during follow-up for BMI, physical activity, regular aspirin/NSAID use and intakes of dietary covariates in regression models.

To evaluate the possible modification by time of the association between fish and PUFA intake and CRC incidence, we constructed various Cox models based on dietary data collected at different time points: (a) Cumulative average intake of fish and PUFAs since baseline through follow-up period to represent the long-term habitual exposure levels; (b) Simple updated intake reported on the most recent FFQ before each follow-up interval; (c) Intake reported at different latencies (i.e., 4-8 years, 8-12 years and 12-16 years) before diagnosis, as described previously³⁵. For example, in the 4-8-year lagged analysis, we related fish and PUFA intake assessed in 1990 to CRC diagnosed between 1994 and 1998. Since our FFQs were administered every four years, the simple updated model could also be considered as a 0-4-year latency analysis.

In subgroup analyses, we assessed whether the associations between fish and PUFA intake and CRC risk varied by selected factors, including aspirin/NSAID use, statin use, postmenopausal hormone use, BMI, physical activity, and fiber intake. We also evaluated potential modification by multivitamin, vitamin E and vitamin C intake as some in vitro studies found that the antioxidant nutrients could abolish the inhibitory effects of fish oil on tumor growth^{36, 37}. A likelihood ratio test was used to compare the model with cross-product terms for the aforementioned factors and primary exposure to that without such terms.

In the secondary analysis, we created prediction models for erythrocyte long-chain ω-3 PUFA by regressing the log transformed erythrocyte EPA and DHA measurements on the dietary intake of dark meat fish, canned tuna fish, and other fish using the FFQs administered in 1990 in the NHS and in 1994 in the HPFS, respectively. Then we applied the cumulative average intake of fish during the entire follow-up period (FFQs in 1984-2006 in the NHS, and 1986-2006 in the HPFS) to these prediction models to estimate the predicted erythrocyte EPA and DHA levels within each cohort, and examined their associations with risk of CRC using the Cox regression model. All statistical tests were two sided and P < 0.05 was considered statistically significant.

Results

Table 1 shows the baseline characteristics of participants by fish intake. The median fish intake was 21.5 g/d (interquartile range: 14.7-37.2 g/d) in women and 29.3 g/d (interquartile range: 14.7-49.9 g/d) in men. In both cohorts, participants with higher fish intake were less likely to be current smokers. They were also more likely to use multivitamins and be physically active, and tended to consume less red meat and more poultry, folate, calcium, vitamin D and dietary fiber.

Table 1.

Baseline age-standardized characteristics of participants according to fish intake (g/d)¹

Characteristic	Nurses’ Health Study				Health Professionals Follow-up Study
	<15	≥15 to <25	≥25 to <40	≥40	<16	≥16 to <30	≥30 to <46	≥46
Participants, %	38.7	23.1	16.7	21.6	33.0	24.2	16.4	26.4
Fish, g/d	9.81	21.2	31.3	65.9	9.95	24.6	37.1	77.3
Age, years	50.3	50.4	50.5	50.5	53.9	53.9	53.9	54.1
Pack-years of smoking before age 30	7.02	6.90	6.92	7.17	11.3	10.8	10.9	11.1
Current smoker, %	26.4	24.5	23.1	21.5	12.1	10.6	9.37	7.57
Family history of colorectal cancer, %	16.8	17.1	17.2	17.0	13.3	13.3	13.9	14.4
History of previous endoscopy, %	4.29	4.56	4.28	4.61	23.8	25.4	26.8	27.8
Current multivitamin use, %	34.3	37.0	38.0	40.7	38.9	40.2	42.3	46.2
Regular aspirin or NSAID use, %2	35.5	35.2	35.7	37.1	30.9	32.5	33.0	34.9
Postmenopausal, %	58.6	58.1	59.2	59.6	-	-	-	-
Current hormone use, %3	23.1	23.1	23.1	22.8	-	-	-	-
Body mass index, kg/m2	24.9	24.8	24.9	25.7	25.0	25.0	24.9	24.8
Physical activity, MET-hours/wk4	12.0	13.7	15.7	17.0	17.8	20.0	23.1	24.6
Dietary intake
Unprocessed red meat, svg/d	0.66	0.65	0.64	0.59	0.69	0.65	0.57	0.50
Processed red meat, svg/d	0.32	0.32	0.32	0.27	0.42	0.40	0.34	0.27
Poultry, svg/d	0.24	0.29	0.33	0.43	0.27	0.33	0.38	0.46
Fruit, svg/d	1.83	2.12	2.39	2.60	1.97	2.26	2.57	2.86
Vegetable, svg/d	2.54	2.97	3.41	3.89	2.51	2.93	3.31	3.78
Alcohol, g/d	6.10	7.44	7.74	6.89	10.7	11.9	12.2	11.2
Folate, μg/d	354	379	399	430	447	464	494	529
Calcium, mg/d	851	866	889	948	897	893	892	906
Vitamin D, IU/d	279	307	336	371	336	373	437	508
Dietary fiber, g/d	15.4	16.2	17.0	17.9	19.4	20.6	21.7	23.2
ω-3 polyunsaturated fatty acids, g/d	1.24	1.32	1.41	1.58	1.20	1.34	1.49	1.73
18:3ω-3 (ALA), g/d	1.04	1.05	1.05	1.08	1.07	1.07	1.08	1.07
Marine ω-3, mg/d	94.0	174	258	412	106	221	339	562
20:5ω-3 (EPA), mg/d	23.6	52.7	82.9	123	32.9	68.3	110	169
22:5ω-3 (DPA), mg/d	12.6	16.9	22.8	32.4	17.1	25.5	36.6	52.5
22:6ω-3 (DHA), mg/d	57.9	105	152	257	56.1	127	192	341
ω-6 polyunsaturated fatty acids, g/d	10.4	10.2	10.1	10.1	11.8	11.6	11.6	11.2
18:2ω-6 (LA), g/d	6.29	6.10	5.98	6.00	11.9	11.7	11.6	11.2
20:4ω-6 (AA), mg/d	119	132	143	168	82.0	86.5	90.3	99.2

¹Continuous variables are described as mean. All variables were assessed in 1984 for the Nurses’ Health Study and in 1986 for the Health Professionals Follow-up Study unless otherwise specified.

²Regular users were defined as those who used ≥2 standard (325-mg) tablets of aspirin or ≥2 tablets of non-steroidal anti-inflammatory drugs (NSAIDs) per week.

³Current hormone use was defined in menopausal women.

⁴Physical activity was measured in 1986 in both cohorts, and represented by the product sum of the metabolic equivalent (MET) of each specific recreational activity and hours spent on that activity per week;

In women, there was no significant association between cumulative average intake of fish, ω-3 and ω-6 PUFAs and overall CRC risk (Table 2). However, fish and marine ω-3 PUFAs appeared to be associated with increased risk of distal colon cancer and a significant heterogeneity was detected by tumor subsite (P_{heterogeneity} = 0.03). The HRs were 1.36 (95% CI: 1.00-1.85; P for trend = 0.05) for ≥40 g/d versus <15 g/d of fish intake and 1.36 (95% CI: 1.03-1.80; P for trend = 0.04) for ≥0.30 g/d versus <0.15 g/d of marine ω-3 intake. When examined individually, the results for PUFAs were similar to the overall estimates. EPA, DHA and DPA were all positively associated with risk of distal colon cancer (Supplementary table 1). We did not observe an association of fish and marine ω-3 PUFAs with proximal colon or rectal cancer.

Table 2.

Hazard ratios (HRs) and 95% confidence intervals (CIs) of colorectal cancer, overall and by subsite, according to cumulative average intake of fish, ω-3 and ω-6 polyunsaturated fatty acids in the Nurses’ Health Study¹

	Colorectal cancer			Proximal colon cancer		Distal colon cancer		Rectal cancer
	Person-years	Cases	HR (95% CI)	Cases	HR (95% CI)	Cases	HR (95% CI)	Cases	HR (95% CI)
Fish intake, g/d
<15	489,799	398	1.00 (referent)	200	1.00 (referent)	97	1.00 (referent)	90	1.00 (referent)
≥15 to <25	486,198	417	1.03 (0.89-1.18)	214	1.02 (0.84-1.24)	109	1.12 (0.85-1.48)	87	0.98 (0.72-1.32)
≥25 to <40	455,394	342	0.91 (0.78-1.06)	163	0.85 (0.68-1.05)	105	1.15 (0.86-1.53)	66	0.79 (0.57-1.11)
≥40	391,099	312	1.02 (0.86-1.20)	136	0.89 (0.70-1.14)	105	1.36 (1.00-1.85)	67	0.98 (0.69-1.40)
P for trend			0.97		0.22		0.05		0.83
18:3ω-3 (ALA), g/d
<0.90	796,473	629	1.00 (referent)	304	1.00 (referent)	169	1.00 (referent)	140	1.00 (referent)
≥0.90 to <1.00	353,691	299	1.10 (0.96-1.27)	149	1.11 (0.90-1.36)	79	1.11 (0.84-1.46)	68	1.19 (0.87-1.61)
≥1.00 to <1.20	435,813	356	1.05 (0.91-1.22)	167	0.99 (0.80-1.22)	112	1.22 (0.93-1.59)	73	1.06 (0.77-1.46)
≥1.20	236,513	185	1.05 (0.86-1.29)	93	1.04 (0.78-1.40)	56	1.18 (0.81-1.71)	29	0.84 (0.52-1.37)
P for trend			0.56		0.89		0.25		0.70
Marine ω-3, g/d
<0.15	660,762	522	1.00 (referent)	266	1.00 (referent)	133	1.00 (referent)	111	1.00 (referent)
≥0.15 to <0.20	307,009	257	1.02 (0.88-1.19)	120	0.93 (0.75-1.15)	76	1.19 (0.89-1.58)	54	1.04 (0.74-1.44)
≥0.20 to <0.30	432,429	352	1.02 (0.89-1.17)	183	1.03 (0.85-1.25)	92	1.05 (0.80-1.39)	73	1.01 (0.74-1.37)
≥0.30	422,290	338	1.03 (0.89-1.20)	144	0.86 (0.69-1.08)	115	1.36 (1.03-1.80)	72	1.06 (0.76-1.48)
P for trend			0.68		0.28		0.04		0.74
Total ω-6, g/d
<8.0	432,787	374	1.00 (referent)	183	1.00 (referent)	100	1.00 (referent)	83	1.00 (referent)
≥8.0 to <9.5	523,013	404	0.91 (0.79-1.06)	203	0.95 (0.77-1.17)	117	0.95 (0.72-1.26)	79	0.83 (0.60-1.15)
≥9.5 to <12.0	623,457	519	1.02 (0.88-1.19)	240	1.02 (0.81-1.27)	150	1.01 (0.76-1.35)	117	1.06 (0.76-1.48)
≥12.0	243,233	172	0.89 (0.70-1.12)	87	1.06 (0.76-1.48)	49	0.74 (0.48-1.15)	31	0.72 (0.43-1.22)
P for trend			0.71		0.64		0.33		0.60

¹Cox proportional hazards model was used and adjusted for age, calendar year, family history of colorectal cancer, prior lower gastrointestinal endoscopy, pack-years of smoking before age 30, body mass index, physical activity, current multivitamin use, postmenopausal status and hormone use, regular aspirin or NSAID use (≥2 tablets/week), total caloric intake, red meat, process meat, alcohol consumption, and energy-adjusted intake of folate, calcium, vitamin D and total fiber.

In men, there was no significant association between fish and PUFA intake and overall risk of CRC (Table 3). However, we did observe that marine ω-3 PUFAs were nonsignificantly associated with risk of distal colon cancer (HR: 1.43; 95% CI: 0.97-2.11 for ≥0.41 g/d versus <0.16 g/d; P for trend = 0.16). Compared to intake of <16 g/d, fish intake of ≥46 g/d was significantly associated with reduced risk of rectal cancer (HR: 0.60; 95% CI: 0.39-0.93; P for trend = 0.05). The P_{heterogeneity} by tumor subsite was 0.28 for fish and 0.41 for marine ω-3 PUFAs. Although the tests for trend did not reach statistical significance, a lower risk of rectal cancer tended to be associated with higher intake of ALA (HR: 0.68; 95% CI: 0.41-1.15 for ≥1.30 g/d versus <0.90 g/d; P for trend = 0.14) and marine ω-3 PUFAs (HR: 0.79; 95% CI: 0.51-1.22 for ≥0.41 g/d versus <0.16 g/d of marine ω-3 intake; P for trend = 0.31). For individual PUFAs, cumulative average intake of EPA, DHA and DPA appeared to have a nonsignificant positive association with distal colon cancer and inverse association with rectal cancer (Supplementary table 2).

Table 3.

Hazard ratios (HRs) and 95% confidence intervals (CIs) of colorectal cancer, overall and by subsite, according to cumulative average intake of fish, ω-3 and ω-6 polyunsaturated fatty acids in the Health Professionals Follow-up Study¹

	Colorectal cancer			Proximal colon cancer		Distal colon cancer		Rectal cancer
	Person-years	Cases	HR (95% CI)	Cases	HR (95% CI)	Cases	HR (95% CI)	Cases	HR (95% CI)
Fish intake, g/d
<16	252,269	264	1.00 (referent)	90	1.00 (referent)	72	1.00 (referent)	68	1.00 (referent)
≥16 to <30	282,051	273	0.90 (0.76-1.07)	91	0.85 (0.63-1.15)	89	1.20 (0.87-1.65)	59	0.73 (0.51-1.04)
≥30 to <46	196,165	226	1.09 (0.90-1.32)	77	1.07 (0.78-1.48)	74	1.45 (1.02-2.05)	43	0.78 (0.52-1.16)
≥46	238,383	224	0.88 (0.72-1.08)	84	0.95 (0.68-1.34)	67	1.12 (0.77-1.64)	45	0.60 (0.39-0.93)
P for trend			0.43		0.90		0.63		0.05
18:3ω-3 (ALA), g/d
<0.90	246,269	272	1.00 (referent)	82	1.00 (referent)	90	1.00 (referent)	59	1.00 (referent)
≥0.90 to <1.10	301,231	299	0.96 (0.80-1.14)	119	1.20 (0.89-1.62)	76	0.82 (0.59-1.13)	70	0.98 (0.68-1.40)
≥1.10 to <1.30	227,070	235	1.01 (0.83-1.23)	88	1.18 (0.85-1.64)	69	1.02 (0.71-1.45)	50	0.89 (0.59-1.35)
≥1.30	194,298	181	0.89 (0.70-1.13)	53	0.81 (0.53-1.23)	66	1.15 (0.75-1.75)	36	0.68 (0.41-1.15)
P for trend			0.45		0.35		0.39		0.14
Marine ω-3, g/d
<0.16	262,869	240	1.00 (referent)	82	1.00 (referent)	65	1.00 (referent)	63	1.00 (referent)
≥0.16 to <0.27	242,720	267	1.18 (0.99-1.41)	93	1.16 (0.86-1.57)	85	1.50 (1.08-2.09)	54	0.90 (0.62-1.31)
≥0.27 to <0.41	228,420	258	1.24 (1.03-1.50)	93	1.31 (0.96-1.79)	80	1.50 (1.06-2.12)	49	0.90 (0.61-1.33)
≥0.41	234,859	222	1.05 (0.85-1.30)	74	1.03 (0.72-1.47)	71	1.43 (0.97-2.11)	49	0.79 (0.51-1.22)
P for trend			0.82		0.95		0.16		0.31
Total ω-6, g/d
<10.0	281,931	304	1.00 (referent)	102	1.00 (referent)	94	1.00 (referent)	64	1.00 (referent)
≥10.0 to <12.0	294,336	278	0.97 (0.81-1.14)	101	1.04 (0.78-1.39)	81	0.93 (0.68-1.26)	61	1.04 (0.72-1.50)
≥12.0 to <14.0	221,201	218	1.05 (0.87-1.27)	76	1.13 (0.82-1.56)	70	1.03 (0.74-1.44)	49	1.14 (0.77-1.70)
≥14.0	171,400	187	1.17 (0.95-1.44)	63	1.25 (0.88-1.79)	56	0.97 (0.66-1.41)	41	1.30 (0.84-2.03)
P for trend			0.12		0.19		0.98		0.23

¹Cox proportional hazards model was used and adjusted for age, calendar year, family history of colorectal cancer, prior lower gastrointestinal endoscopy, pack-years of smoking before age 30, body mass index, physical activity, current multivitamin use, regular aspirin or NSAID use (≥2 tablets/week), total caloric intake, red meat, process meat, alcohol consumption, and energy-adjusted intake of folate, calcium, vitamin D and total fiber.

Tables 4 and 5 present the associations between marine ω-3 PUFA intake and overall and subsite-specific risk of CRC in varying lag-time periods among women and men, respectively. In women, no significant association was detected or appeared to be modified by time (Table 4). In men, there was no association between marine ω-3 and CRC for the 0-4-year, 4-8-year or 8-12-year lag (Table 5). However, for the 12-16-year lag, we observed a nonsignificant inverse association (HR: 0.76; 95% CI: 0.52-1.10 comparing extreme categories; P for trend = 0.15). By subsite, marine ω-3 PUFAs tended to be more strongly associated with reduced risk of rectal cancers with longer latency. When extreme categories were compared, the HRs for rectal cancer were 1.04 (95% CI: 0.68-1.59) for the 0-4-year lag and 0.68 (95% CI: 0.33-1.43) for the 12-16-year lag, respectively.

Table 4.

Hazard ratios (HRs) and 95% confidence intervals (CIs) of colorectal cancer, overall and by subsite, according to intake of marine ω-3 polyunsaturated fatty acids with variable latency periods in the Nurses’ Health Study¹

	No. of cases	HR (95% CI)				P for trend
		<0.15 g/d	≥0.15 to <0.20 g/d	≥0.20 to <0.30 g/d	≥0.30 g/d
Colorectal cancer
0-4-year lag	1,295	1.00 (referent)	0.91 (0.75-1.10)	1.11 (0.96-1.29)	0.97 (0.84-1.13)	0.93
4-8-year lag	874	1.00 (referent)	0.79 (0.62-1.01)	1.09 (0.91-1.31)	0.96 (0.80-1.15)	0.86
8-12-year lag	650	1.00 (referent)	0.98 (0.76-1.26)	1.11 (0.90-1.37)	0.87 (0.69-1.11)	0.41
12-16-year lag	508	1.00 (referent)	0.92 (0.69-1.21)	0.90 (0.71-1.14)	0.98 (0.77-1.26)	0.79
Proximal colon cancer
0-4-year lag	630	1.00 (referent)	0.87 (0.66-1.15)	1.16 (0.94-1.44)	0.83 (0.66-1.03)	0.19
4-8-year lag	435	1.00 (referent)	0.73 (0.50-1.04)	1.13 (0.87-1.46)	0.86 (0.66-1.13)	0.44
8-12-year lag	330	1.00 (referent)	0.90 (0.63-1.29)	1.20 (0.90-1.60)	0.74 (0.52-1.06)	0.23
12-16-year lag	266	1.00 (referent)	1.10 (0.75-1.60)	1.12 (0.81-1.54)	1.01 (0.71-1.44)	0.87
Distal colon cancer
0-4-year lag	366	1.00 (referent)	1.03 (0.73-1.44)	1.05 (0.79-1.41)	1.06 (0.80-1.40)	0.68
4-8-year lag	244	1.00 (referent)	0.81 (0.52-1.27)	1.02 (0.71-1.45)	0.93 (0.66-1.32)	0.78
8-12-year lag	170	1.00 (referent)	1.26 (0.80-1.97)	0.95 (0.61-1.46)	0.86 (0.54-1.37)	0.50
12-16-year lag	132	1.00 (referent)	0.64 (0.34-1.18)	0.66 (0.40-1.10)	1.02 (0.64-1.63)	0.88
Rectal cancer
0-4-year lag	277	1.00 (referent)	0.84 (0.55-1.28)	1.14 (0.82-1.58)	1.18 (0.86-1.61)	0.26
4-8-year lag	185	1.00 (referent)	0.95 (0.57-1.58)	1.09 (0.73-1.64)	1.17 (0.79-1.73)	0.41
8-12-year lag	145	1.00 (referent)	0.88 (0.50-1.55)	1.11 (0.71-1.74)	1.15 (0.72-1.82)	0.51
12-16-year lag	105	1.00 (referent)	0.94 (0.52-1.71)	0.73 (0.42-1.28)	0.92 (0.53-1.57)	0.62

¹Cox proportional hazards models were used and adjusted for the same sets of covariates as in Table 2.

Table 5.

Hazard ratios (HRs) and 95% confidence intervals (CIs) of colorectal cancer, overall and by subsite, according to intake of marine ω-3 polyunsaturated fatty acids with variable latency periods in the Health Professionals Follow-up Study¹

	No. of cases	HR (95% CI)				P for trend
		<0.16 g/d	≥0.16 to <0.27 g/d	≥0.27 to <0.41 g/d	≥0.41 g/d
Colorectal cancer
0-4-year lag	847	1.00 (referent)	1.03 (0.84-1.26)	1.21 (1.00-1.46)	1.06 (0.87-1.30)	0.53
4-8-year lag	569	1.00 (referent)	1.05 (0.81-1.35)	1.09 (0.87-1.37)	0.92 (0.71-1.18)	0.46
8-12-year lag	363	1.00 (referent)	1.05 (0.78-1.41)	0.96 (0.71-1.28)	0.90 (0.66-1.24)	0.48
12-16-year lag	253	1.00 (referent)	0.85 (0.60-1.20)	0.86 (0.61-1.20)	0.76 (0.52-1.10)	0.15
Proximal colon cancer
0-4-year lag	307	1.00 (referent)	0.86 (0.60-1.21)	1.20 (0.88-1.63)	1.12 (0.80-1.56)	0.38
4-8-year lag	200	1.00 (referent)	0.81 (0.52-1.25)	0.89 (0.61-1.31)	0.81 (0.54-1.23)	0.39
8-12-year lag	133	1.00 (referent)	0.78 (0.48-1.28)	0.62 (0.37-1.02)	0.72 (0.43-1.20)	0.17
12-16-year lag	94	1.00 (referent)	1.00 (0.57-1.75)	0.86 (0.49-1.51)	0.69 (0.35-1.33)	0.25
Distal colon cancer
0-4-year lag	263	1.00 (referent)	1.24 (0.87-1.76)	1.30 (0.93-1.82)	1.02 (0.70-1.49)	0.96
4-8-year lag	187	1.00 (referent)	1.21 (0.78-1.89)	1.61 (1.09-2.39)	0.79 (0.49-1.28)	0.34
8-12-year lag	113	1.00 (referent)	1.94 (1.13-3.33)	1.80 (1.06-3.06)	1.35 (0.73-2.48)	0.43
12-16-year lag	72	1.00 (referent)	0.68 (0.33-1.40)	1.07 (0.58-1.94)	0.80 (0.39-1.65)	0.71
Rectal cancer
0-4-year lag	186	1.00 (referent)	0.98 (0.63-1.52)	1.17 (0.78-1.75)	1.04 (0.68-1.59)	0.82
4-8-year lag	121	1.00 (referent)	1.18 (0.69-2.01)	0.96 (0.58-1.61)	0.99 (0.58-1.69)	0.87
8-12-year lag	85	1.00 (referent)	0.91 (0.49-1.68)	0.90 (0.50-1.62)	0.69 (0.35-1.35)	0.28
12-16-year lag	65	1.00 (referent)	1.06 (0.54-2.06)	0.79 (0.39-1.61)	0.68 (0.33-1.43)	0.27

¹Cox proportional hazards models were used and adjusted for the same sets of covariates as in Table 3.

We considered the possibility that long-term intake may confound the association of more recent intake and therefore simultaneously evaluated marine ω-3 PUFAs consumed within the immediately preceding 10 years and that consumed more than 10 years in the past. High marine ω-3 intake within immediately preceding 10 years was not associated with lower overall or subsite-specific risk of CRC after controlling for intake more than 10 years in the past (data not shown). However, high intake greater than 10 years in the past was associated with progressively lower risk of rectal cancer in men, even after adjusting for marine ω-3 consumption within the immediately preceding 10 years (HR for comparing extreme categories: 0.48; 95% CI: 0.24-0.96; P for trend = 0.07).

Since ALA and LA compete for desaturases and elongases for conversion into long-chain PUFAs, we examined the ratio of dietary ALA to LA in relation to CRC risk. In women, we did not find any significant association with CRC either overall or by subsite (data not shown). By contrast, the ratio was associated with reduced risk of CRC, proximal colon cancer and rectal cancer in men. Compared the ALA:LA ratio of ≥0.12 to <0.09, HRs were 0.80 (95% CI: 0.61-1.04; P for trend = 0.04) for overall CRC, 0.59 (95% CI: 0.37-0.96; P for trend = 0.01) for proximal colon cancer, and 0.74 (95% CI: 0.42-1.32; P for trend = 0.14) for rectal cancer. We did not find any significant association between the ratio of marine ω-3 to total ω-6 PUFAs, the ratio of total ω-3 to ω-6 PUFAs and CRC risk (data not shown).

We evaluated the potential modification of the marine ω-3 PUFA-CRC relationship by aspirin/NSAID use, statin use, postmenopausal hormone use, BMI, physical activity, and consumption of fiber, multivitamin, vitamin E and vitamin C. No significant heterogeneity was detected by any of these factors (P for heterogeneity > 0.05).

In the secondary analyses using predicted erythrocyte EPA and DHA concentrations, we generally observed similar results as dietary measurements (Supplementary Table 3). Among women, a positive association with risk of distal colon cancer was observed comparing extreme quartiles of predicted erythrocyte EPA (HR: 1.38; 95% CI: 1.02-1.86; P for trend = 0.02) and DHA (HR: 1.52; 95% CI: 1.12-2.05; P for trend = 0.002). Among men, an inverse association was observed with rectal cancer comparing extreme quartiles of predicted erythrocyte EPA (HR: 0.62, 0.40-0.94; P for trend = 0.04), for EPA and DHA (HR: 0.61; 95% CI: 0.40-0.93; P for trend = 0.04).

Discussion

In two large U.S. prospective studies, we did not observe an overall association between fish or PUFA intake and CRC risk, which is consistent with several other large prospective studies in which no association was found between ω-3 PUFA intake and CRC risk^{17, 38-40}. This is in contrast to laboratory evidence demonstrating antineoplastic activity of ω-3 PUFAs. Possible explanations for the discrepant findings include the considerably higher dose of ω-3 PUFAs used in laboratory studies compared to general human consumption, variation in timing of exposure to ω-3 PUFAs in tumorigenesis, inherent difference in biological effects of short-term exposure compared with long-term intake, and the influence of additional factors that may be present in human cohorts but absent in experimental models².

Despite these overall null results, we found significant associations between marine ω-3 PUFAs and some specific subsites of colorectal tumors. Participants with high fish or marine ω-3 intake appeared to have an associated increased risk of distal colon cancer, whereas marine ω-3 intake was inversely related to risk of rectal cancer in men after long-term follow-up. Previous studies have recognized the potential for substantial variations in the biochemical environment throughout the large bowel which may contribute to etiologic heterogeneity in carcinogenesis⁴¹. Bacterial numbers, fermentation and proliferation are highest in the proximal colon. Thus, the production of short-chain fatty acid (SCFA) from fermentation also varies within the intestinal tract, with the total amount of SCFA ranging from 70 to 140 mM in the proximal colon and falling to 20 to 70 mM in the distal colon, which leads to the gradual elevation in the pH from proximal to distal colon⁴². Therefore, high intake of ω-3 PUFA may protect against cancer development in the proximal colon mucosa, while elevated pH milieu may attenuate or even reverse the potential benefit of ω-3 PUFA on distal colon cancer through unidentified mechanisms. Alternatively, contaminants present in some fish species, such as mercury, dioxins and polychlorinated biphenyls (PCBs), may contribute to the increased risk of distal colon cancer. The carcinogenic effect of methylmercury, inorganic mercuric chloride, dioxins, PCBs has been suggested in animal studies, although the human evidence is inadequate or limited^{43, 44}. A prospective study in Shanghai reported an increased risk of CRC among women with high intake of eel, shrimp, and shellfish, which are mainly raised in industrial areas and enriched with chemicals with potential carcinogenicity⁴⁵. Consistent with our findings, a large prospective study in Japan also reported an increased risk of distal colon cancer among individuals with high ω-3 PUFA intake²⁰. To our knowledge, other data examining CRC by subsites beyond the simple dichotomous classification of colon and rectal cancer are lacking. Thus, further studies based on more specific anatomic subsites, which have also been related to differences in molecular characteristics, are needed.

Our observation of an inverse association between marine ω-3 PUFA intake and rectal cancer is in line with the anti-inflammatory hypothesis and experimental evidence. It has been shown that there is a notable overexpression of prostaglandin-endoperoxide synthase 2 (PTGS2, also known as cyclooxygenase-2, COX-2) protein in tumors located in the rectum, compared with other locations in the colon⁴⁶. PTGS2 catalyzes the conversion of AA to eicosanoids, namely prostaglandins and leukotrienes, and is a crucial mediator in the link between inflammation and carcinogenesis. High ω-3 PUFA suppresses the biosynthesis of AA-derived eicosanoids through either direct inhibition of PTGS2 activity or competition with ω-6 for PTGS to form anti-inflammatory eicosanoids². Thus, it is biologically plausible that ω-3 PUFA may be more beneficial for prevention of rectal cancer than colon cancer through its actions in suppression of PTGS2. In agreement with this hypothesis, regular use of aspirin, which also inhibits PTGS2 at high doses, has also been more strongly associated with lower risk of rectal cancer than colon cancer in both observational studies⁴⁷ and randomized trials⁴⁸. Notably, a recent phase III randomized double-blind trial investigated treatment with enteric-coated formulation of EPA as the free fatty acid (EPA-FFA) 2g daily for 6 months in patients with familial adenomatous polyposis who had previously undergone colectomy with ileorectal anastomosis. They found that in the treatment group the number of rectal polyps was significantly reduced by 22.4% and the sum of polyp diameters decreased by 29.8% compared with the placebo group⁴⁹, a similar magnitude of reduction to those seen with the selective PTGS2 inhibitor celecoxib⁵⁰, thus highlighting the promise of EPA-FFA as a chemoprevention agent for rectal cancer.

In addition, a large European study also found that high intake of fish was associated with a greater reduction in risk of rectal cancer compared to colon cancer⁴. A Japanese study observed a U-shaped pattern between marine ω-3 PUFA intake and rectal cancer: compared to men with the lowest intake (median intake = 0.49), those at the second and third quintiles (median intake = 0.79 and 1.06 g/d) had a 31% and 49% reduced risk of rectal cancer, respectively; but those at the fifth quintiles (median intake = 2.18 g/d) had a somewhat increased risk²⁰. Despite the dramatic difference in intake between our participants with those in the Japanese study, this relationship is in fact broadly consistent with our findings that an intake range of 0.16-0.41 g/d was associated with 10-21% reduced risk of rectal cancer compared to intake of <0.16 g/d. These results suggest that an intake of 0.30-1.00 g/d of marine ω-3 PUFA may be favorable for rectal cancer prevention, whereas intake beyond this level may not. However, given the limited data, more studies are needed to determine the optimal dose of marine ω-3 PUFA for cancer prevention.

By examining the latency between marine ω-3 PUFA intake and CRC incidence, we found that the inverse association did not emerge until after 12-16 years of follow-up in men. This finding is consistent with the results of some prior studies. In the Physicians’ Health Study, the investigators observed that the significant association between baseline ω-3 PUFA intake and reduced CRC risk among men was restricted to cases occurring after follow-up of 13 years¹⁸. Likewise, the Singapore Chinese Health study reported that among participants with ≤5 years of follow-up high marine ω-3 PUFA intake at baseline had a significantly positive association with advanced CRC; by contrast, among those followed for at least 10 years a somewhat inverse association was observed¹⁷. Given the potential modification by time, it is possible that variation in the length of follow-up might partially explain the overall inconsistent findings in previous prospective studies^{14, 17, 18, 20, 38-40}. In addition, because colorectal carcinogenesis is characterized by a relatively slow and stepwise progression from normal to dysplastic epithelium to carcinoma, which may take years or even decades, our findings of a long latency are consistent with marine ω-3 PUFAs having a greater influence on initiation or early promotion of incident neoplasia rather than on progression of established tumors. This is also in accordance with the necessity of a minimum of 5-10 years of aspirin use on CRC prevention⁴⁸, since both aspirin and marine ω-3 PUFA may act against the inflammatory pathway, which represents an important early event in the initiation of carcinogenesis. However, the findings for adenoma, the precursor lesion of CRC, have been mixed. Some^{51, 52} but not all⁵³ case-control studies observed an inverse association between dietary intake of ω-3 PUFA and adenoma risk. A prospective study using the NHS data suggested an inverse association of higher marine ω-3 PUFA intake with large distal colorectal adenoma⁵⁴. Considering the sparse data, more prospective studies, particularly in men, are still needed.

As in other studies^{38, 46}, we did not find any significant relation for ω-6 PUFA, marine ω-3/ ω-6 ratio, or total ω-3/ ω-6 ratio. Interestingly, we detected a significant association between higher ALA/LA ratio and lower risk of CRC, specifically proximal colon cancer, in men. ALA, a non-marine source of ω-3 PUFA, competes with LA for desaturases and elongases for conversion to more bioactive long-chain PUFAs. It is possible that high ALA intake in combination with low LA intake favors the formation of EPA and DHA over that of AA, thus contributing to an antineoplastic effect in the colorectal mucosa. However, this was a subset analysis limited to men and could be due to chance.

We observed some suggestion of sex difference in the PUFA-CRC relationship. Consistent with prior studies^{38, 55} and the recent meta-analysis¹⁴, the association of marine ω-3 with CRC seemed to be more pronounced in men than in women. Estrogen may alter the normal FA metabolism through changes in FA utilization and oxidation. On the other hand, intake of PUFA may also affect endogenous estrogen production, as AA-derived prostaglandin E₂ stimulates the activity of aromatase P450². Although estrogen may represent a potential explanation for the observed sex difference in marine ω-3 and CRC risk, we did not detect any significant modification by postmenopausal hormone use. Thus further studies examining the modification by endogenous estrogen levels are needed. With regard to other possible modifiers, we did not find any notable heterogeneity in the subgroup analyses.

There are several limitations of our study. First, dietary assessment by the FFQ is subject to measurement errors that have several sources, ranging from report error, different nutrient content of foods to between-subject variation in absorption and metabolism of PUFA not captured by the questionnaire. Nevertheless, the association between dietary intake calculated from the FFQ and multiple biomarkers of PUFA and similar results from secondary analyses using predicted erythrocyte PUFA levels are reassuring. Second, although we have adjusted for a variety of potential confounders and performed mutual adjustment for ω-3 and ω-6 analysis, residual confounding by shared dietary and lifestyle pattern could not be excluded due to the possibility of measurement error and model misspecification. However, the homogeneous population of health professionals makes it unlikely that residual confounding explains our results. Third, the modest consumption and variation of fish and ω-3 PUFA in our study population limits the ability to examine the dose-response relationship across a wide exposure range, especially at the high end of intake. The ongoing VITamin D and OmegA-3 TriaL (VITAL) with treatment of marine omega-3 fatty acid of 1 g/d may provide additional information in this regard, but long-term follow-up might be needed given our findings of latency analysis. Finally, a large number of comparisons and statistical tests were conducted in our analyses. Thus some findings may occur by chance and should be interpreted with caution.

The strengths of the current study included the prospective design with over 20 years follow-up, high follow-up rate, large sample size, detailed data on many established risk factors for CRC, and informative analyses by tumor subsite. Uniquely, the repeated measurement of fish and PUFA intake enabled us to estimate long-term consumption with greater accuracy, and to perform detailed latency analyses for the identification of the most critical period of action.

In conclusion, this prospective study does not support a strong association between fish, ω-3 and ω-6 PUFA intake and overall CRC risk. However, some findings suggested in our study deserve further research, including the association of marine ω-3 PUFAs with increased risk of distal colon cancer and reduced risk of rectal cancer among men, and the potential modification by time and sex of the association between marine ω-3 PUFA intake and CRC.

novelty and impact.

Through comprehensive analyses of two large cohorts, our findings suggest no association between fish, ω-3 and ω-6 polyunsaturated fatty acid intake and overall colorectal cancer incidence. However, high consumption of marine ω-3 might be associated with increased risk of distal colon cancer, but reduced risk of rectal cancer. Using the unique data of repeated dietary assessments over long-term follow-up, our latency analysis suggests a long induction period for protection of marine ω-3 against colorectal cancer.

Abbreviations used

PUFA

polyunsaturated fatty acid

CRC

colorectal cancer

HR

hazard ratio

CI

confidence interval

ALA

α-linolenic acid

EPA

eicosapentaenoic acid

DHA

docosahexaenoic acid

DPA

docosapentaenoic acid

LA

linoleic acid

AA

arachidonic acid

NHS

Nurses’ Health Study

HPFS

Health Professionals Follow-up Study

FFQ

food frequency questionnaire

NSAIDs

non-steroidal anti-inflammatory drugs

MET

metabolic equivalent

BMI

body mass index

SCFA

short-chain fatty acid

PTGS2

prostaglandin-endoperoxide synthase 2

VITAL

VITamin D and OmegA-3 TriaL