Alcohol intake and colorectal cancer risk: A dose–response meta-analysis of published cohort studies

Abstract

The epidemiologic evidence support that alcohol intake might be associated with increased colorectal cancer risk. However, the results by anatomic site in the large bowel are inconsistent. We conducted a meta-analysis of prospective cohort studies published between 1990 and June 2005 on the relationship between alcohol intake and colon and rectal cancer. We quantified associations with colon and rectal cancer using meta-analysis of relative risk (RR) associated to the highest versus the lowest category of alcohol intake and meta-analysis of study-specific dose–response slopes using fixed or random effect models depending on the heterogeneity of effects among studies. Sixteen prospective cohort studies including more than 6,300 patients with colorectal cancer were eligible for inclusion. High alcohol intake was significantly associated with increased risk of colon (RR = 1.50; 95% CI = 1.25, 1.79) and rectal cancer (RR = 1.63; 95% CI = 1.35, 1.97) when comparing the highest with the lowest category of alcohol intake, equivalent to a 15% increase of risk of colon or rectal cancer for an increase of 100 g of alcohol intake per week. The relationship did not differ significantly by anatomical site (colon, rectum). Using meta-regression analysis, we identified geographical area where the study was conducted as a possible source of between-study heterogeneity of effects among studies. Lifestyle recommendations for prevention of colorectal cancer should consider limiting alcohol intake. © 2006 Wiley-Liss, Inc.

Alcohol is widely consumed throughout the world and is thought to be related to more than 60 different medical conditions.1 For most diseases, there is a dose–response relation to volume of alcohol consumption, with risk of disease increasing with higher volume. One exception is coronary heart disease, for which a J-shaped curve is described. The risk relation reverses for higher average volume of alcohol consumption.2 Other dimensions rather than total volume, especially irregular heavy drinking, have been shown to play a crucial role in some entities, in particular injuries.3

Alcohol consumption is one of the most important known causes of human cancer.4 Several studies suggest that increased alcohol is a risk factor for colorectal cancer. Previous reviews5, 6, 7, 8 and meta-analyses9, 10, 11 of case–control and cohort studies suggested that high alcohol intake might be associated with an increased risk of colorectal cancer.4 The epidemiological evidence has been complemented by recent molecular evidence on mechanisms that could explain the association.4

However, the findings on the association of alcohol intake and the risk of colorectal cancer by anatomical site in the large bowel are still inconsistent. The results of the most recent review suggest that the association could be stronger with rectal compared to that of colon tumors,10 but in a recent pooled analysis of 8 cohort studies in European and North American countries,12 no difference in association by anatomical site in the large bowel was observed. Moreover, the results of at least 8 larger prospective cohort studies have been published with divergent results in Asian and Western countries.13, 14, 15, 16, 17, 18, 19, 20

Determining whether alcohol increases risk of colon and rectal cancer is important because a substantial number of individuals drink alcohol, and alcohol consumption is increasing rapidly in many parts of the world.20 Moreover, colorectal cancer is one of the most common cancers in Western countries.21 In some areas, such as in Japan and elsewhere in Asia, its incidence rate has markedly increased over the past 40 years.21 These trends have been generally ascribed to changes in diet. Alcohol intake also grew rapidly. The alcohol intake has been common in men and has also become more popular in women. Because drinking is a potential risk factor for colorectal cancer, this increase in consumption may partly explain the increase in incidence and mortality rates of this malignancy.20 Modifying alcohol consumption could be part of a prevention strategy of colorectal cancer through lifestyle changes. The aim of this study was to examine if current alcohol intake is associated with risk of colon and rectal cancer by summarizing the results of published prospective cohort studies with meta-analytic techniques.

Materials and methods

Identification of prospective cohort studies

We identified publications in MEDLINE using alcohol, colorectal cancer, colon, rectum, cohort, case–cohort or nested case–control as keywords. We also examined the references from the identified articles and previous reviews on alcohol and cancer. The criteria for inclusion were (i) prospective cohort studies evaluating the relationship between total alcohol consumption and colorectal cancer risk; (ii) published in English between 1990 and June 2005; (iii) referenced in MEDLINE; (iv) with colorectal cancer incidence as endpoint; (v) providing relative risk (RR) estimates and its corresponding 95% confidence intervals (95% CI), or information allowing us to compute unadjusted variance.22 To be included in the dose–response meta-analyses, studies had to report associations for at least 3 categories of exposure, number of cases and comparison subjects for each category. Nested case–control studies or case–cohort studies carried out within well-defined cohorts were also included. Since studies on specific types of alcohol (beer, wine, liquor) were limited, we restricted the meta-analyses to total alcohol consumption on colorectal cancer risk. Studies in particular populations (i.e., cohorts of alcoholics or brewery workers) were not included.

Statistical methods

We extracted the maximally adjusted RRs and CIs. Odd-ratios were considered an estimate of RRs for nested case–control studies. We first quantified the association of alcohol with colorectal cancer risk as the weighted mean of the logarithm of RR estimates associated to the highest versus the lowest category of alcohol consumption reported in each study.22 Second, we estimated study-specific dose–response slopes by relating the logarithm of the RRs for different exposure levels to their corresponding alcohol content, using the method described by Greenland and coworkers23, 24 We expressed results as RRs for an increase of 100 g/week of alcohol, corresponding approximately to 5 and 7 drinks per week in Asia and USA respectively.16, 25, 26 Five studies did not provide the number of person-years by levels of intake.13, 14, 18, 27, 28 It was approximated from follow-up duration and number of subjects. Most studies provided alcohol consumption in grams or ounces.13, 15, 20, 27, 28, 29, 30 We rescaled alcohol consumption when reported as number of drinks by multiplying the number of drinks by using standard alcohol content per drink in the same study population14, 16, 18, 25, 31 or the geographical area.16, 17, 19, 20, 26, 31 We assigned as exposure value to each category the mid-point of the corresponding range of alcohol intake. When the highest category was open-ended, we assumed the width of the interval to be the same as in the preceding category. Random effect models were assumed when there was evidence of heterogeneity. We quantified the extent of heterogeneity using Q-test32 and I² score.33 We used meta-regression to explore the influence of tumor site, gender, geographical region, dose of ethanol intake in the highest category (<100, 100–200, ≥200 g/week) and publication year in the heterogeneity.24 Publication bias was assessed using the tests of Egger34 and Begg and Mazumbar.35 All statistical analyses were done with Stata Statistical Software, version 8.0.

Results

We identified 16 cohort studies that evaluated the association of colon, rectum or colorectal cancer incidence and alcohol intake. Five cohort studies reported RRs for colorectal cancer,13, 16, 18, 19, 28 14 for colon cancer14, 15, 16, 17, 19, 20, 26, 27, 28, 30, 36, 37, 38 and 12 for rectal cancer14, 15, 16, 19, 20, 26, 27, 28, 30, 37, 38 (Table I). Three cohorts published results in two different articles each27, 29, 31, 37, 38, 39 whereas one article presented results of two different cohorts.38 One study30 reported sex-specific RRs but not CIs, nor the information to compute them. The RR for both sexes combined was included. We calculated unadjusted CIs for one study.26 For one study,28 we included results from the nonsupplemented carotene group because the positive association between level of alcohol consumption and colorectal cancer risk was attenuated in subjects supplemented with β-carotene.

Table I. Study Characteristics of Published Prospective Cohort Studies on Alcohol Intake and Colorectal Cancer Risk

Author(s) and year	Name of the cohort or description	Country (follow-up period)	Sex	Sites explored in analyses	No. of cases	Number of subjects1	Number of categories of exposure2	Range of exposure	Variables of adjustment	Type of questionnaire	Age at baseline	Mean age
FFQ, Food-Frequency Questionnaire; Q, Questionnaire; CC, Colorectal Cancer; M, men; W, women. 1 Number of subjects in the whole cohort, controls for nested-case–control studies or number of subject in the subcohort for case–cohort studies. 2 Current alcohol intake.
Wakai et al. (2005)20	Japan Collaborative Cohort Study	Japan (1988–1997)	M/W	Colon	418	57,736	5 (men); 3 (women)	0–153 vs. ≥462 g/week	Age, area, education, family history of CC, BMI, smoking habits, walking time sedentary work, consumption of green leafy vegetables and beef	Self-reported Lifestyle questionnaire	40–79	–
				Rectum	211
Chen et al. (2005)19	Population-based cohort, Jiashan County	China (1990–2001)	M/W	Colorectal	242	64,100	3	Nondrinker vs. ≥ 7 drinks/week	Age, sex, smoking status, occupation, education level, marital status	Face-to-face Q	30 and more	–
				Colon	107
				Rectum	135
Sanjoaquin et al. (2004)18	Oxford Vegetarian Study	United Kingdom (1980–1999)	M/W	Colorectal	95	10,998	3	<1 vs. >7 units/week	Age, sex, smoking status	Self-reported FFQ	16–89	–
Su and Arab (2004)17	NHANES (NHEFS)	United States (1982–1993)	M/W	Colon	111	10,418	3	Nondrinker vs. ≥1 drink/day	Age, sex, smoking status, race, BMI, Educational level, nonpoultry meat consumption, poultry and seafood consumption, multivitamin use, history of colonic polyps	Interview, FFQ	25–74	–
Wei (2004)38	Nurses' Health Study (NHS)	United States (1980–2000)	W	Colon	672	87,733	4	Nondrinker vs. ≥20 g/day	Age, family history, BMI, physical activity, beef, pork or lamb as a main dish, processed meat, alcohol, calcium, folate, height, pack-year smoking before age 30, history of endoscopy	FFQ	30–55	–
				Rectum	204
Wei (2004)38	Health Professionals Follow-up Study (HPFS)	United States (1986–2000)	M	Colon	467	46,632	4	Nondrinker vs. ≥20 g/day	Age, family history, BMI, physical activity, beef, pork or lamb as a main dish, processed meat, alcohol, calcium, folate, height, pack-year smoking before age 30, history of endoscopy	FFQ	40–75	–
				Rectum	135
Otani (2003)16	Japan Pulic Health Center-based propective study on cancer (JPHC)	Japan (1990–1999)	M/W	Colorectal	706	90,004	5 (men); 3 (women)	Nondrinker vs. ≥300 g/week (≥1 g/week women)	Age, family history of colorectal cancer, BMI, smoking status, physical activity, geographic areas	Self-reported Q	40–69	–
				Colon	299 (men)
				Rectum	259 (men)
Shimizu (2003)15	cohort of Japaneseresidents in Takayama City	Japan (1993–2000)	M/W	Colon	198	29,051	3	Nondrinker vs. >36.7 g/day	Age, height, BMI, smoking and education	Self-reported semi-FFQ	–	54.1 men, 55.1 women
				Rectum	97
Pedersen (2003)14	Danish population-based cohort, Copenhagen	Denmark (1964–1999)	M/W	Colon	411	29,132	6	<1 vs. ≥41 drinks/week	Age, sex, smoking, BMI, study of origin	Self-reported Q	23–95	–
				Rectum	202
Harnack (2002)37	Iowa Women's Health Study (IWHS)	United States (1986–1998)	W	Colon	598	41,836	2	<20 vs. ≥ 20 g/day	Age, smoking, BMI, estrogens use, calcium, vitamin E, energy	Self-reported Q	–	61.7
				Rectum	123
Flood (2002)13	Breast Cancer Detection Demonstration Project (BCDDP)	United States (1979–1998)	W	Colorectal	490	45,264	5	Nondrinker vs. >2 serving/day	Energy, dietary folate, methionine, smoking	Self-reported FFQ	40–93	61.9
Singh (1998)36	Adventist Health Study	United States (1976–1983)	M/W	Colon	157	32,051	2	<1 vs.≥1 times/week	Age, sex, parental history of colon cancer	Self-reported semi-FFQ	25 and more	–
Chyou (1996)27	Japanese Hawaii Cancer Study	United States (1965–1995)	M	Colon	330	7945	4	Nondrinker vs. ≥24 oz/mo	Age	Face-to-face Q	45 and more	–
				Rectum	123
Glynn (1996)28	α-Tocopherol β-carotene Cancer Prevention Study(ATBC)	Finland (1985–1993)	M	Colorectal, colon	87	27,109	4	0.01–5.3 vs. >27.7 g/day	Age, physical activity during work, intakes of total energy, starch, sweet and sugar, coffee, calcium	Self-reported Food use Q	50–69	–
				Rectum	53
Murata (1996)26	Nested-case-control study in the Gastric Mass survey program	Japan (1984–1993)	M	Colon	61	61/122	4	Nondrinker vs. ≥2.1 cups/day	Smoking status	Self-reported Q	–	–
				Rectum	43	43/86
Giovannucci (1995)31	Health Professionals Follow-up Study (HPFS)	United States (1986–1992)	M	Colon	205	47,931	5	0–0.25 vs. >2 drinks/day	Age, history of polyps, history of endoscopy, parental history of colorectal cancer, smoking, BMI, physical activity, aspirin use, energy, red meat, methionine and folate	Self-reportedsemi-FFQ	40–75	–
Gapstur (1994)29	Iowa Women's Health Study (IWHS)	United States (1986–1990)	W	Colon	237	38,006	3	Nondrinkervs. ≥4 g/day	Age	Self-reported Q	55–69	–
				Rectum	75
Goldbohm (1994)30	Case-cohort study in the Netherlands Cohort Study (NLCS)	Netherlands (1986–1989)	M/W	Colon	299	3170	5	0.1–4.9 vs. ≥30 g/day	Age, smoking, Quetelet index, history of gallbladder surgery, level of education, intake of energy and energy-adjusted intake of fat, meat protein, and dietary fibre	Self-reported Q	55–69	–
				Rectum	153
Stemmermann (1990)39	Japanese Hawai Cancer Study	United States (1965–1989)	M	Colon	211	7572	5	Nondrinkervs. ≥40 oz/mo	Age, smoking, years of smoking	Face-to-face Q	–	–
				Rectum	101

In Table II we present the overall RR of colorectal, colon and rectal cancer comparing the highest category of alcohol intake versus the lowest. Two studies reported a significant increased risk of colorectal cancer only among men16, 28 and the remaining studies did not report significant associations.13, 16, 18, 19 Overall, alcohol intake was positively but not significantly associated with colorectal cancer (RR_{highest vs. lowest category} = 1.34; 95% CI = 0.92, 1.96). The results were heterogeneous across cohorts (p Het = 0.002, I² = 71%). High alcohol intake was significantly associated with colorectal cancer in men (RR_{highest vs. lowest category} = 1.73; 95% CI =1.00, 2.98, p Het = 0.02, I² = 74%) but not in women.

Table II. Average RR of Colorectal Cancer by Anatomical Site for the Highest Versus the Lowest Level of Intake of Alcohol

Subgroup	Colorectal			Colon			Rectum
	Number of datasets1	RR (95% CI)	p Het. (I^2 score2)	Number of datasets1	RR (95% CI)	p Het. (I^2 score2)	Number of datasets^1	RR (95% CI)	p Het.(I^2 score2)
1 Studies provided separated analyses for men and women were analysed as two separates cohorts. 2 Percentage of total variation across studies that is due to heterogeneity rather than chance.33
All studies	713, 16, 18, 19, 28	1.34 (0.92, 1.96)	0.002 (71%)	1714, 15, 16, 17, 19, 20, 26, 27, 28, 30, 36, 37, 38	1.50 (1.25, 1.79)	0.03 (43%)	1414, 15, 16, 19, 20, 26, 27, 28, 30, 37, 38	1.63 (1.35, 1.97)	0.68 (0%)
Men	316, 19, 28	1.73 (1.00, 2.98)	0.02 (74%)	815, 16, 19, 20, 26, 27, 28, 38	1.64 (1.39, 1.93)	0.06 (48%)	715, 16, 20, 26, 27, 28, 38	1.79 (1.38, 2.33)	0.43 (0%)
Women	313, 16, 19	0.88 (0.61, 1.27)	0.44 (0%)	515, 19, 20, 37, 38	1.23 (1.00, 1.51)	0.43 (0%)	415, 20, 37, 38	1.39 (0.95, 2.02)	0.72 (0%)
Both sexes	118	1.53 (0.87, 2.69)	–	414, 17, 30, 36	1.33 (0.97, 1.83)	0.13 (47%)	314, 19, 30	1.54 (1.00, 2.37)	0.46 (0%)
Asia	416, 19	1.16 (0.64, 2.13)	<0.001 (83%)	815, 16, 19, 20, 26	1.80 (1.45, 2.24)	0.18 (32%)	715, 16, 19, 20, 26	1.61 (1.22, 2.13)	0.63 (0%)
Europe	218, 28	1.83 (1.14, 2.92)	0.27 (18%)	314, 28, 30	1.39 (0.54, 3.57)	0.04 (68%)	314, 28, 30	2.04 (1.09, 3.82)	0.91 (0%)
USA	113	1.16 (0.63, 2.14)	–	617, 27, 36, 37, 38	1.34 (1.15, 1.56)	0.4 (2%)	427, 37, 38	1.57 (1.18, 2.11)	0.17 (40%)

In analysis by cancer site, alcohol was significantly associated with colon cancer (RR_{highest vs. lowest category} =1.50; 95% CI =1.25, 1.79). The results were heterogeneous (p Het = 0.03, I² = 43%; Fig. 1). Ten cohorts studies reported significant positive associations,15, 16, 17, 20, 26, 27, 28, 36, 38 5 studies reported positive but nonsignificant relationships19, 20, 30, 37, 38 and 2 studies14, 19 reported nonsignificant inverse relationships between alcohol intake and colon cancer incidence. Alcohol intake was positively associated to colon cancer in men (RR_{highest vs. lowest category} = 1.64; 95% CI = 1.39, 1.93) but not in women. Alcohol was significantly positively associated with rectal cancer (RR_{highest vs. lowest category} = 1.63; 95% CI = 1.35, 1.97; Fig. 2). There was no significant heterogeneity. Out of 12 studies on rectal cancer, 3 studies reported significant associations14, 16, 27 and the remaining studies found a positive but nonsignificant relationship.15, 19, 20, 26, 28, 30, 38 One study37 conducted on women reported a nonsignificant inverse association. The association with cancer of the rectum was significant in men (RR_{highest vs. lowest category} = 1.79; 95% CI =1.38, 2.33) but not in women. The positive association with rectal cancer was consistent in all geographical areas.

Figure 1. RRs (highest vs. lowest category) for published cohort studies or nested case–control studies on the between total alcohol consumption and colon cancer incidence. Black squares show the study-specific RRs and 95% CI. The size of the black squares reflects the study-specific weight in the pooled analysis. Diamonds represent the combined RRs and 95% CI. NC, Non Consumer.

Figure 2. RRs (highest vs. lowest category) for published cohort studies or nested case–control studies on the between total alcohol consumption and rectal cancer incidence. Black squares show the study-specific RRs and 95% CI. The size of the black squares reflects the study-specific weight in the pooled analysis. Diamonds represent the combined RRs and 95% CI. NC, Non Consumer.

In meta-regression analyses, gender was close to statistical significance for colorectal (p = 0.07) and colon cancer (p = 0.06), but the relationship was attenuated after inclusion of alcohol level intake in the model. Geographical area was a significant source of heterogeneity between studies for colon cancer (p = 0.02) and remained significant after adjustment for level of alcohol intake (p = 0.03).

We did not find evidence of publication bias either with the Egger's test (p = 0.35 for colorectal cancer, p = 0.19 for colon cancer, p = 0.70 for rectal cancer) or Begg's test (p = 0.88 for colorectal cancer, p = 0.25 for colon cancer, p = 0.34 for rectal cancer).

We included 14 cohort studies in the dose–response analysis (Table III). We excluded 2 studies with only 2 categories of alcohol consumption36, 37 and one publication with 2 cohorts38 that did not provide number of person years separately for each cohort. The 3 excluded studies reported a positive relationship between alcohol intake and colorectal cancer, which was significant in one study.36 We included the results for both sexes combined for one study19 because the distribution of cases by sex was not provided. There was a 19% increase of colorectal cancer associated to an increase of 100 g of alcohol per week (RR= 1.19; 95% CI = 1.14, 1.27). The RR of colon cancer was 1.15 for a 100 g/week increase in alcohol intake. The latter results were heterogeneous (p < 0.001). The dose response relationship between alcohol intake and colon cancer risk was significant in men (RR for a 100 g/week increase =1.18; 95% CI = 1.13, 1.24) and women (RR = 1.14; 95% CI = 1.00, 1.30). The RR of rectal cancer was 1.15 for a 100 g/week increase in alcohol intake. There was no significant heterogeneity. The dose–response relationship was statistically significant in men (RR for a 100 g/week increase = 1.19; 95% CI = 1.12, 1.26), but not in women. The results were consistent in all geographic areas. The dose–response relationship for rectal cancer was of similar magnitude to the value observed for colon cancer (Table IV).

Table III. RRs (95% CIs) of Colorectal Cancer for an Increase of 100 g of Alcohol Intake Per Week by Anatomical Site

Subgroup	Colorectal			Colon			Rectum
	Number of datasets1	RR (CI 95%)	p Het.	Number of datasets1	RR (CI 95%)	p Het.	Number of datasets1	RR (CI 95%)	p Het.
1 Studies provided separated analyses for men and women were analysed as two separates cohorts.
All studies	713, 16, 18, 19, 28	1.19 (1.14, 1.27)	0.11	1414, 15, 16, 17, 19, 20, 26, 27, 28, 29, 30, 31	1.15 (1.07, 1.23)	<0.001	1214, 15, 16, 19, 20, 26, 27, 28, 29, 30	1.15 (1.10, 1.21)	0.61
Men	316, 19, 28	1.21 (1.02, 1.43)	0.05	715, 16, 20, 26, 27, 28, 31	1.18 (1.13, 1.24)	0.16	615, 16, 20, 26, 27, 28	1.19 (1.12, 1.26)	0.29
Women	313, 16, 19, 28	1.05 (0.92, 1.20)	0.84	315, 20, 29	1.14 (1.00, 1.30)	0.16	315, 20, 29	1.16 (0.94, 1.44)	0.79
Both sexes	118	1.24 (0.76, 2.01)	–	414, 17, 19, 30	0.99 (0.94, 1.05)	0.2	314, 19, 30	1.10 (1.02, 1.19)	0.9
Asia	416, 19, 28	1.21 (1.14, 1.27)	0.22	715, 16, 19, 20, 26	1.15 (1.10, 1.21)	0.46	715, 16, 19, 20, 26	1.16 (1.09, 1.23)	0.87
Europe	218, 28	1.44 (1.10, 1.87)	0.48	314, 28, 30	1.14 (0.85, 1.52)	0.02	314, 28, 30	1.10 (1.02, 1.20)	0.8
USA	113	1.02 (0.87, 1.20)	–	417, 27, 29, 31	1.23 (1.12, 1.35)	0.17	227, 29	1.43 (1.18, 1.72)	0.76

Table IV. RRs (95% CI) for Selected Doses of Alcohol Intake Per Week on Colon and Rectal Cancers

Cancer site	RR (CI 95%)
	25 g/week	50 g/week	100 g/week
Colon	1.03 (1.02, 1.05)	1.07 (1.03, 1.11)	1.15 (1.07, 1.23)
Rectum	1.04 (1.02, 1.05)	1.07 (1.05, 1.10)	1.15 (1.10, 1.21)

In meta-regression analysis, geographical area was identified as a possible source of heterogeneity for colon after adjustment of alcohol level intake (p = 0.003).

We further repeated the analysis including only the 10 studies14, 15, 16, 19, 20, 26, 27, 28, 29, 30 that provide results for both colon and rectal cancer in order to better compare results by anatomical site. We removed 2 studies providing only results for colon cancer.17, 36 The dose–response relationship was similar to that observed in the unrestricted analysis (RR for a 100 g/week increase = 1.14; 95% CI = 1.09, 1.20) for colon and rectum site (Table III).

Discussion

We have systematically reviewed published prospective cohort studies on the association between total alcohol and the risk of colon and rectal cancer. Higher alcohol intake was associated with increased risk of colon and rectal cancer. This quantitative summary of published prospective cohort studies is consistent with previous reviews10, 11, 12 and supports the linear relationship between alcohol intake and risk of colon and rectal cancer.10 The overall RRs estimated in this study are comparable to the values reported by the Pooling projects of cohort studies.12 We included in our meta-analyses 11 cohort studies that were not included in the pooling project.13, 14, 15, 16, 17, 18, 19, 20, 26, 27, 36 No significant difference between RRs of colon and rectal cancer emerged in relation to alcohol intake. Similar observations were found in previous meta-analyses11 and the Pooling Project of cohort studies.12

The results of the studies included in this analysis were heterogeneous. The significant source of heterogeneity identified by meta-regression was geographical area where the study was conducted. Although our results suggest a positive association between alcohol intake and colorectal cancer risk in both men and women, stronger associations were reported in men compared to women, possibly because alcohol intake is higher in men than in women. Alcohol intake in the highest category in each study varied between 140 and 257 g/week in women, while in men the highest intakes ranged from 140 to 462 g/week approximately. However, the dose-response relationships were similar in men and in women. In meta-regression analyses, the influence of gender in the heterogeneity disappeared after adjustment for alcohol level intake. The significance of geographical area as a possible source of heterogeneity remained after adjustment for alcohol level intake. Highest levels of alcohol intake in the studies ranged from 154 to 462 g/week in Asian studies, 15–159 g/week in American studies and 194–492 g/week in European studies. Level of alcohol in the highest category and geographical area are correlated making difficult to identify their own specific effects.

Furthermore, evidence suggests that the effect of alcohol is modulated by polymorphisms in genes encoding enzymes for ethanol metabolism (e.g., alcohol dehydrogenases, adehyde dehydrogenases). The mechanisms by which alcohol consumption exerts its carcinogenic effect have not been defined fully, although plausible events include a genotoxic effect of acetaldehyde, the main metabolite of ethanol.4 Murata et al. showed that colorectal cancer risk was enhanced for heterozygous carriers of a specific allele of the ALDH2 gene.40 ALDH2*2 is frequently found in Asian populations, whereas nearly all Europeans are homozygous for the ALDH2*1 allele.4 One study41 suggested that polymorphism in the ADH3 may modify the alcohol-colorectal cancer risk association.

Our study has several strengths. We included all the cohort studies published so far for a total of 6,300 colorectal cancer cases. Prospective studies have the advantage of being less vulnerable to selection and recall bias than case–control studies. We were able to evaluate the associations for colon and rectal cancer separately and for each gender. Other than considering the overall RR for the highest compared to the lowest level of alcohol intake provided in the epidemiological literature, we assessed the dose–response relationship. As a meta-analysis of published studies, we are exposed to publication bias although our results seem to suggest that there was no evidence of publication bias.

In our meta-analysis, results are combined from studies that vary in study populations, methods of assessment of alcohol intake, levels and type of alcohol consumed. In the estimation of the average RR associated with the highest versus the lowest category of alcohol consumption, we rescaled exposure to a comparative scale to estimate the dose–response relationship. Adjustment for potential confounders varied across studies. All studies included in our meta-analysis provided RR estimates adjusted for main risk factors.

We did not investigate alcohol intake according to type of alcoholic beverage or drinking pattern. A recent study12 did not find any difference in risk by type of alcoholic beverage, suggesting that the positive association is attributable to ethanol intake itself rather than to a specific beverage. Previous evidence does not support that one source of alcohol could be more strongly associated with colorectal cancer risk than alcohol as a whole.8, 42

Measures of drinking patterns have not been included in most epidemiologic studies. In addition, no standardized measure of drinking pattern has been adopted, making comparisons and pooling of studies difficult.1 Both volume of alcohol consumption and patterns of drinking have been shown to influence alcohol-related burden of disease.1

Low-to-moderate alcohol consumption has been shown to reduce the risk of vascular disease by about a third and reduce total mortality in middle and old age.43 The proportion of cancer deaths attributable to alcohol intake worldwide has been estimated as 6%.44 Strategies of prevention of colorectal cancer should consider that the results of this meta-analysis show that risk of colon and rectal cancer is increased by about 50% in subjects with the highest alcohol consumption compared to lower consumers.