Soy Consumption and Colorectal Cancer Risk in Humans: A Meta-Analysis

The role of diet in the etiology of colorectal cancer, the third leading cause of death by cancer in both men and women in the United States (1), remains an area of active research. The age-standardized incidence rate of colorectal cancer in Asian countries (e.g., China) is lower than that in North America and European countries (2), and the incidence increases substantially in migrants from low-risk to high-risk areas (3, 4). These observations indicate that differences in lifestyle, including dietary practices, between the Eastern and the Western populations may play a role in colorectal cancer etiology.

Soy is a major plant source of dietary protein for humans. An early review of epidemiologic studies (most of which were case-control studies published before 2000) suggested an inverse association between high soy intake and colon cancer risk in humans (5). In recent years, studies with relatively large study populations specifically designed to assess soy consumption in association with colorectal cancer risk have been conducted, some of which showed a significant inverse association (6, 7). Furthermore, soy diets inhibit chemically induced colon tumorigenesis in animals (8, 9), and isoflavones, a group of bioactive components in soy, may play a role in this inhibitory process (10). There are hardly any clinical studies of the relation between soy and colorectal cancer (11, 12).

The purpose of the present study was to conduct meta-analyses of currently available epidemiologic studies of the association between consumption of soy and colorectal cancer risk to provide a quantitative evaluation in a standardized format that permitted a numerical analysis across studies.

We conducted a Medline search, supplemented with a hand-search of article bibliographies and non-indexed medical and professional journals reporting epidemiologic studies that provided quantitative data on soy and colorectal cancer. We used the following terms in combination for the literature search: soy (tofu, soymilk, miso, natto), isoflavones (genistein, daidzein), colorectal cancer (colon, rectum), and epidemiology (cohort, case-control). In addition, we conducted a broader search on diet and colorectal cancer aimed at identifying studies in which the aforementioned terms were not included in abstracts. Furthermore, we contacted investigators for unpublished results that might be useful for the analysis. The search was conducted through May 31, 2009. We systematically reviewed and examined whether the identified studies met the following criteria to be included in the analysis: a study (cohort or case-control) must have had soy consumption assessed as a food and/or isoflavone consumption assessed from food intake; it must have provided a risk estimate [hazard ratio, relative risk, or odds ratio (OR)] for colorectal, colon, or rectal cancer incidence as well as its 95% confidence interval (95% CI); and it must have provided information on adjustment for confounding factors. From the results of these studies, we extracted the risk estimate of the highest relative to the lowest intake for the analysis.

The hazard ratio and relative risk were taken to be approximations to OR, and the meta-analysis was done as if all types of ratio were odds ratios. This was justified by the fact that colorectal cancer had a relatively low incidence over the periods of time studied. We calculated the combined risk estimate using a random-effects model in which the effect measures were log OR–weighed by the method of DerSimonian and Laird (13), giving greater weight in the summary measure to studies with smaller standard error of estimate. Furthermore, this model incorporated a test of heterogeneity (Q statistic) into its calculation by which it estimates the magnitude of the heterogeneity and assigns a greater variability to the estimate of overall treatment effect to account for this heterogeneity. We used the methods of Begg and Mazumdar (14) and Egger et al. (15) to detect publication bias. Both methods test for funnel plot asymmetry, the former (14) being based on the rank correlation between the effect estimates and their sampling variances, and the latter (15) on a linear regression of a standard normal deviate on its precision. If a potential bias was detected, we further conducted a sensitivity analysis to assess the robustness of combined effect estimates and the possible influence of the bias and to have the bias corrected. We used the Stata version 9.2 statistical program (StataCorp) for the analysis. All reported P values are from two-sided statistical tests, and differences with P ≤ 0.05 were considered significant.

We identified 11 studies on soy consumption and colorectal cancer incidence that met the inclusion criteria (Table 1). Four of these studies were cohort studies (6, 7, 16, 17), and seven were case-control studies (18-24). Furthermore, we identified four studies that assessed soy intake and colorectal cancer mortality (Table 1; refs. 25-28). Food frequency questionnaire was the method of dietary assessment in all of these studies. Because different types of soy foods were evaluated in these studies and some assessed more than one type of soy food, we chose the risk estimate of a measurement in these studies that was the most representative of overall soy consumption or a soy food item that was the most commonly consumed for the analysis. These measurements were prioritized in descending order of total soy foods or soy products, tofu (bean curd) or miso soup (soy paste soup). We also identified three cohort (6, 7, 17) and three case-control studies (29-31) that assessed isoflavone intake in association with colorectal cancer incidence (Table 2). Seven studies separately presented findings for men and women (7, 17, 19, 22, 25, 26, 28), and one study separately reported results for risk of proximal and distal colon cancer (22). We included data from both men and women and from both proximal and distal colon cancer as independent populations in this analysis. Four studies were excluded from this analysis, one that assessed soy together with other legumes as one food group (32), one that evaluated soy together with vegetables and fruits as a dietary pattern (33), and two that provided no information on confounding factors (34, 35).

Our analysis of the 11 studies that assessed soy and colorectal cancer incidence (6, 7, 16-24) yielded a combined risk estimate of 0.90 (95% CI, 0.79-1.03; P = 0.134) and test of heterogeneity Q = 35.75 (P = 0.011; Fig. 1). The results of our analysis of the four studies on soy and colorectal cancer mortality (25-28) showed a combined risk estimate of 0.98 (95% CI, 0.85-1.14; P = 0.790), with Q = 14.63 (P = 0.102). No publication bias was detected in either analysis by the tests of Begg and Mazumdar (14) and Egger et al. (15), respectively.

Of the 11 studies we analyzed, six separately provided data on women (6, 7, 16, 17, 19, 22) and four on men (7, 17, 19, 22). Our analysis of the six studies on women yielded a combined risk estimate of 0.79 (95% CI, 0.65-0.97; P = 0.026) with Q = 10.92 (P = 0.142; Fig. 2A), and that of the four studies on men yielded a combined risk estimate of 1.10 (95% CI, 0.90-1.33; P = 0.358) with Q = 6.51 (P = 0.260; Fig. 2B). Publication bias was detected in neither analysis.

We further separately analyzed studies that separately provided data on colon cancer and rectal cancer. Seven studies provided data on colon cancer (6, 7, 17, 20, 22-24), and our analysis of these studies yielded a combined risk estimate of 0.88 (95% CI, 0.74-1.06; P = 0.175) with Q = 20.81 (P = 0.035). Five studies provided data on rectal cancer (6, 17, 20, 22, 23), and our analysis of these studies yielded a combined risk estimate of 0.88 (95% CI, 0.67-1.14; P = 0.329) with Q = 12.32 (P = 0.055). No publication bias was detected in either analysis.

Because there were differences in study types (cohort or case-control), study populations (Asians or non-Asians), and types of soy foods assessed (soy foods or products, tofu or miso) among the studies we analyzed, we further conducted stratified analyses to determine the impact of these influences on our analysis. The results of these stratified analyses are in Table 3. Noteworthy, for the category of soy foods/products (6, 7, 17, 23) and of miso (17, 18, 20, 22-24), combined risk estimates were found of 0.79 (95% CI, 0.62-1.00; P = 0.047) and 0.89 (95% CI, 0.82-0.97; P = 0.008), respectively, whereas the combined risk estimate for tofu (16, 18-22) was not significant (0.97; 95% CI, 0.83-1.13; P = 0.689), all without publication bias being detected.

The results of the analysis of the six studies on isoflavones (6, 7, 17, 29-31) showed a combined risk estimate of 0.76 (95% CI, 0.51-1.13; P = 0.173) with Q = 146.06 (P = 0.001) and P = 0.711 and P = 0.034 for publication bias by the methods of Begg and Mazumder (14) and Egger et al. (15), respectively. A sensitivity analysis excluding the Ravasco et al. study (31), which caused asymmetry of the funnel plot, yielded a combined risk estimate of 0.84 (95% CI, 0.72-0.98; P = 0.025) with Q = 11.61 (P = 0.071; Fig. 3) and P = 0.133 and P = 0.116 for publication bias by the Begg and Mazumder (14) and Egger et al. tests (15), respectively. We further carried out stratified analyses on the basis of study sites, because three of these studies were conducted in Asia (6, 7, 17) and the other three in Western countries with non-Asians (29-31). The results of these analyses, presented in Table 3, were suggestive of a reduced risk in the non-Asian populations, but not in the studies of Asian populations.

In the present study, we analyzed 11 epidemiologic studies (6, 7, 16-24) that assessed the association of soy consumption with colorectal cancer incidence in humans. In two of these studies soy intake was found to be associated with a significant reduction in colorectal cancer risk in women (6, 7), whereas in the other nine studies no such association was observed in either women or men. Our combined analysis of these 11 studies showed that consumption of soy foods is not associated with the risk of colorectal cancer or colon or rectal cancer separately, and the results from our analysis of the four studies on soy and colorectal cancer mortality (25-28) support this finding.

In separate analyses of studies that provided data on the basis of gender, we found that soy consumption was associated with an approximately 21% reduction in colorectal cancer risk in women. This finding may be related to the structural and metabolic similarities of soy isoflavones to mammalian estrogens because of the following observations by others. Epidemiologic (36) and clinical studies (37) have shown a significant reduction in colorectal cancer risk in postmenopausal women who used hormone replacement therapies. In human samples, estrogen receptor gene expression was diminished or absent in colorectal tumors, and introduction of an exogenous estrogen receptor gene in cultured colon carcinoma cells resulted in marked growth suppression (38). Dietary supplementation with isoflavones increased estrogen receptor–β expression but reduced estrogen receptor–α expression in colon of female rats (39), and feeding soy protein to ovariectomized rats in combination with estrone treatment resulted in a greater inhibition of azoxymethane-induced colon tumorigenesis than soy protein alone (40).

We did not find that soy consumption was associated with colorectal cancer risk in men. There is strong evidence from laboratory studies that dietary supplementation with soy protein (8), soy flakes or soy flour (9), fermented miso (41), or soybean curd refuse (42) inhibits experimentally induced colon tumorigenesis in male rats, and soy isoflavones seem to play a role in this inhibition (10). The possible mechanisms include a decrease in colon fatty acid synthesis (43), a reduction in polyamine production (44), an induction of somatostatin (45), and an increase in fecal fat excretion (46). However, human clinical studies are very limited. Although one small-scale intervention study showed that isoflavone-containing soy protein reduces crypt cell proliferation in colon mucosa biopsies from both male and female subjects with a history of colon polyps or colon cancer (11), another study did not show such a protective effect in men and women (12). Obviously, more clinical trials are warranted to investigate the role of soy intervention in colorectal cancer prevention in both genders.

Our analysis of six studies on isoflavones (6, 7, 17, 29-31) showed that isoflavone consumption was associated with an approximately 16% reduction in colorectal cancer risk. This significant risk reduction was largely attributable to three studies conducted in Western countries (29-31), which was also reflected in the results of the stratified analysis. However, in a recently completed study in the United Kingdom neither serum isoflavones (OR, 1.01; 95% CI, 0.94-1.08) nor urinary isoflavones (OR, 1.03; 95% CI, 0.95-1.11) were associated with colorectal cancer risk (47). Of the six studies we analyzed, Rossi et al. (30) and Cotterchio et al. (29) assessed isoflavone intakes at low μg/d to low mg/d levels and reported a significant reduction in colorectal cancer risk. In contrast, all three studies conducted in Asian countries (6, 7, 17) reported intakes at levels in the range of several mg/d to >50 mg/d, and none of those showed a significant risk reduction. Furthermore, all of the studies conducted in Western countries (29-31) were case-control studies with relatively small study populations, and none of them were designed to study soy. Thus, large-scale investigations specifically designed to study soy are warranted to understand the possible protective role of soy isoflavones in colorectal cancer development in Western populations.

We conducted stratified analyses of the 11 studies on soy and colorectal cancer risk to determine the impact of differences in study types, study populations, and types of soy foods assessed on our analysis. Results from the analyses of studies by study types (cohort versus case-control) and by study populations (Asians versus non-Asians) were very similar to that from our analysis of the 11 studies. Of the 11 studies analyzed, 4 provided data on soy foods or soy products (6, 7, 17, 23), 6 on tofu (16, 18-22), and 6 on miso (17, 18, 20, 22-24). Results from the stratified analyses showed that soy foods/products and miso intakes were associated with a significant reduction in colorectal cancer risk, whereas tofu consumption was not associated with colorectal cancer risk. Both tofu and miso are the most commonly consumed soy foods in Asian countries, and many epidemiologic studies have used tofu (16, 25) or miso (27) as a marker of soy consumption. Of the four studies that reported intake of soy foods or soy products (6, 7, 17, 23), three included tofu (6, 7, 17) in the category of soy foods or soy products. Thus, caution should be taken in interpreting these results.

The present study provides a quantitative analysis of available epidemiologic studies on soy consumption and colorectal cancer risk in humans. We conducted stratified analyses, examined publication bias and had it corrected if detected, and used the random-effects model (13), which estimates the magnitude of the heterogeneity and assigns a greater variability to the combined risk estimate to account for the heterogeneity if detected. Like all meta-analyses, however, our analysis may have limitations. First, most of the studies, particularly the early case-control studies, were not designed solely to evaluate the association between soy consumption and colorectal cancer risk, and there were wide variations in dietary assessments among the studies. For example, of the 11 studies, only five reported the use of a validated questionnaire (6, 7, 16, 17, 19) whereas the other studies did not provide such information. Most of the studies assessed frequency of intake, but only a few evaluated intake quantity, and the measurement units varied across the studies. Second, the studies varied in the number of potential confounding factors for which they had adjusted. Some cohort studies published in recent years provided detailed information of adjustment for confounders (6, 16, 17), whereas some early case-control studies adjusted for fewer such factors (18, 22). Third, the use of food frequency questionnaires in case-control studies, in which recall bias is an inherent problem, may have affected the result of the analysis.

In summary, our analysis of currently available epidemiologic studies suggests that consumption of soy foods is associated with a reduction in colorectal cancer risk in women, but not in men. Considering that colorectal cancer is the third leading cause of death by cancer in both men and women in the United States (1), further investigations with strong dietary assessment tools and appropriate adjustment for confounding factors are warranted to evaluate the potentially protective role of dietary soy and isoflavones in colorectal cancer etiology.

None of the authors had any conflicts of interest to report.

We gratefully acknowledge Dr. Lu Wang, Harvard University, for kindly providing us the unpublished data to this analysis and the assistance of Mr. Jay Dahlgran, Grand Forks Human Nutrition Research Center, in preparing tables presented in the manuscript.

Grant Support: USDA ARS research project 5450-51000-036-00D.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.