Introduction

Diet plays an important role in the etiology of certain cancers; however, establishing precisely which dietary factors account for this influence has proved to be difficult. Certain food groups such as vegetables and fruits contain phytochemicals with anti-carcinogenic properties. Soy foods and soybean components, in recent years, have particularly received considerable attention for their potential role in reducing cancer risk. Although the breast and prostate cancer has been the focus of most interest, there is an expanding body of literature on the possible association between soy foods and gastrointestinal (GI) cancers.

A number of anti-carcinogenic phytochemicals in soybeans have been identified, including phytosterols, phenolic acids and protease inhibitors [1]. Among these, most attention has been focused on the component isoflavones. Isoflavones have a limited distribution in nature with soybeans being its main source [2]. Although many investigators have focused on its anti-estrogenic properties and its potential for preventing hormonally mediated cancers, they also act as antioxidants and possess other anti-carcinogenic activities, including inhibition of angiogenesis [3], topoisomerase [4] and tyrosine kinase [5].

Genistein, one of the predominant isoflavones in soy, has been demonstrated to inhibit proliferation of cultured cells, including normal and transformed intestinal epithelial cells in a number of rodent and human cancer cell lines [6].

While the mechanisms that underpin an inverse relation between soy consumption and GI cancer risk seem plausible, the evidence from epidemiological data to date has been inconsistent and controversial. A recent meta-analysis examining four cohort and nine case–control studies on soy consumption and colorectal cancer (CRC) risk found a 21 % risk reduction in women; however, the overall analysis showed no significant association [7]. Two separate meta-analyses reviewed the effects of soy foods on gastric cancer risk on different population groups [1, 8]. Pooled results from their meta-analyses suggested that the risk of stomach cancer may depend on whether the soy food was fermented or non-fermented, with non-fermented products being associated with a greater risk reduction. Results from in vitro and in vivo animal studies have also been inconsistent with regard to a protective effect of soy and colorectal cancers. Five studies showed a statistically significant effect of soy diet or isoflavone supplement inhibiting the formation of aberrant crypt foci in rats, which are preneoplastic lesions of CRC [9]. Three studies that tested soy protein isolate on chemically induced colon cancer did not note a reduction in carcinogenesis [10–12].

Previous epidemiological studies have used different parameters in reporting the serving sizes and have included different types of soy in their analysis, making interpretation of results difficult. Gender influence has also been an area of interest given the anti-estrogenic properties of soy isoflavone. This systematic review and meta-analysis aim to summarize epidemiological findings to explore and clarify the relationship between soy intake and GI cancer, including subgroup analyses of dietary isoflavone intake, anatomical subsites, the influence of gender and types of soy food consumed.

Methods

Study protocol

Literature searches of epidemiological studies in this systematic review were performed using the meta-analysis of observational studies in epidemiology (MOOSE) guidelines where possible [13]. The following electronic databases were searched: MEDLINE, PubMed, ISI Web of Science, Current Contents Connect and Embase. The search included all studies published up to May 27, 2014. Key terms including ‘Soy’ OR ‘Isoflavone’ AND ‘Gastrointestinal neoplasm’ were searched as text words and as exploded medical subject headings where possible. References in the relevant review articles from the bibliographic database search were also checked for appropriate studies. No language restrictions were used in either the search or study selection. A search for unpublished literature was not performed.

Study selection

The following inclusion criteria were applied in the screening of articles: (1) original data on soy consumption and GI neoplasms risk, that of the esophagus, stomach and/or colorectum, were provided; (2) the risk point estimate was reported as OR or RR, or the data were presented such that an OR could be calculated; (3) the 95 % confidence interval (CI) was reported, or the data were presented such that the CI could be calculated.

Data extraction

Data extraction was performed via a standardized data extraction form, collecting information on the publication year, study design, number of cases and controls, total sample size, temporal direction, population type, country, ethnicity of sample group, case–control matching, mean age, response rate, exposure, neoplasm type, number of adjusted variables, method of intake measurement, and the risk estimates or data used to calculate the risk estimates, CIs. Quality of the studies was not assessed, and authors were not contacted for missing data. Adjusted odds ratios (ORs) were extracted in preference to non-adjusted odds ratios; however, where OR was not provided, unadjusted ORs and CIs were calculated. Where more than one adjusted ratio was reported, the ratio with the highest number of adjusted variables was chosen. Where multiple risk estimates were available in the same study, for example, due to the use of mutually exclusive comparator groups, they were included as separate risk estimates. Where ORs were provided in tertiles, quartiles or quintiles, the middle tiles were included.

Statistical analysis

Pooled OR and 95 % confidence intervals were calculated for the effect of soy intake on the risk of GI neoplasms using a random effects model, model of DerSimmonian and Laird [14]. Heterogeneity with Cochran’s Q statistic was tested, with p < 0.10 indicating heterogeneity, and the degree of heterogeneity was quantified using the I 2 statistic, which represents the percentage of the total variability across studies due to heterogeneity. I 2 values of 25, 50 and 75 % corresponded to low, moderate and high degrees of heterogeneity, respectively [15]. Publication bias was quantified using the Egger’s regression model [16]. All analyses were performed with comprehensive meta-analysis (version 2.0).

Effect modification was tested between soy intake and other covariates (country, gender, GI site, specific soy foods and method of soy preparation) through the addition of multiplicative interaction terms into the model. A separate analysis for dietary isoflavone intake and GI cancer risk was performed. As different methods were used to report soy intake, sensitivity analysis on serving sizes involved categorizing studies using a standardized measurement-consumption of three or more times per week and excluding those that did not specify intake frequency. Analyses were performed comparing studies that used validated versus non-validated FFQs.

Results

The literature searches identified 348 articles for evaluation. Title and abstract screening excluded 292 articles due to duplicates, non-human or non-original research. Full text screening excluded 15 articles—seven studies did not include soy or isoflavone in the dietary assessment, confidence intervals could not be calculated in six studies, and two studies did not present original data. Twenty-two case–control and 18 cohort studies were included in the final meta-analysis, which contained a total of 633,476 participants and 13,639 GI cancer cases. Characteristics of studies are outlined in Table 1.

Table 1 Characteristics of included case–control and cohort studies reporting odd ratios and 95 % CIs for the association between soy/isoflavone intake and GI neoplasm risk
Full size table

Eighteen case–controls and 16 cohort studies provided data on soy intake. The combined OR was consistent with an inverse association between soy consumption and GI cancer, calculated to be 0.93 (95 % CI 0.87–0.99; p value heterogeneity = 0.01), showing only a slight decrease in risk. Results for both case–control (OR 0.86; 95 % CI 0.72–1.02); p value heterogeneity = 0.07) and cohort studies (OR 0.97; 95 % CI 0.90–1.03; p value heterogeneity = 0.21) were similar. Analysis of variables including country, gender, study design, type of soy consumed and cancer type was conducted (Table 2). Egger’s regression analysis for assessment of publication bias of the studies included in this meta-analysis found no significant bias (p < 0.001) (Fig. 1).

Table 2 Odd ratios with 95 % CIs for variables examined for the association between soy intake and gastrointestinal neoplasm risk
Full size table
Fig. 1
figure 1

Funnel plot showing studies assessed for publication bias of soy intake and GI cancer risk

Full size image

When analyzing according to GI sites, colon cancer (OR 0.92; 95 % CI 0.96–0.99; p value heterogeneity = 0.163) and CRC (OR 0.92; 95 % CI 0.87–0.97; p value heterogeneity = 0.3) showed a weak but statistically significant reduction in risk. Gastric cancer and rectal cancer were associated with a small reduction in risk as well; however, the results were not significant. Only two studies examined the effects of soy intake on esophageal cancer risk and hence were not included in the analysis.

Seven case–control and 12 cohort studies specifically examined gender variation on the influence of soy intake on GI cancer risk. Our analysis showed little difference between men and women. There is a modest but statistically significant inverse association for both men (OR 0.93; 95 % CI 0.87–0.99; p value heterogeneity = 0.01) and women (OR 0.92; 95 % CI 0.85–0.99; p value heterogeneity = 0.03). Subgroup site-specific analysis showed that the association was statistically significant for colon cancer in women (OR 0.85; 95 % CI 0.73–0.99; p value heterogeneity = 0.4).

Of the 41 included studies, only seven were from Western countries and the rest were of Asian origin. Studies from China and Japan showed a reduced risk of GI cancers with ORs of 0.81 (95 % CI 0.74–0.87; p value heterogeneity < 0.001) and 0.95 (95 % CI 0.92–0.99; p value heterogeneity = 0.03), respectively.

Results for fermented and non-fermented soy products were similar, yielding ORs of 0.97 (95 % CI 0.88–1.06; p value heterogeneity = 49.13) and 1.03 (95 % CI 0.92–1.15; p value heterogeneity = 0.104), respectively. Due to a large number of studies from Asian countries assessing tofu (bean curd) and miso (soybean paste soup) intake, we further conducted stratified analysis to determine the effects of specific soy products. Our results showed no significant association or difference between intake of either soy product and GI cancer risk.

Five case–control and five cohort studies included the subgroup analysis on dietary isoflavone intake (Fig. 2). Combined analysis yielded a statistically significant inverse association with a risk estimate of 0.73 (95 % CI 0.59–0.92; p value heterogeneity = 0). Site-specific evaluation showed a significant inverse association with CRC (OR 0.76; 95 % CI 0.59–0.98; p value heterogeneity = 0) only. Only five studies provided separate data on men and women; our analysis yielded very similar results for gender.

Fig. 2
figure 2

Forest plot of isoflavone intake and GI cancer risk

Full size image

Of the studies included in our analysis, only five case–control and seven cohort studies employed a validated food frequency questionnaire (FFQ) to measure soy intake. Sensitivity analysis that excluded non-validated FFQs reported a risk estimate of 0.95 (95 % CI 0.882–1.01; p value heterogeneity = 0.004), yielding similar results for when all studies were included.

Discussion

Our results show that dietary isoflavone intake is strongly associated with a decrease in risk of developing GI cancers. Evidence for the consumption of soy and GI cancer intake has not been as consistent, showing only a small decrease in risk. To our knowledge, this is the first systematic review to investigate soy and dietary isoflavone intake in relation to GI cancer risk, taking into account the outcome of different anatomical subsites, and in influence of gender differences and other covariants.

All but one study included in our analysis of the effects isoflavone intake have been published within the past decade. Results from seven out of the ten studies support the hypothesis that isoflavone is inversely associated with GI cancer risk. Soybeans and soy products contain roughly 1–3 mg isoflavones per gram protein. Isoflavones are also known as ‘phytoestrogens’ because of the structural and metabolic similarities to mammalian estrogens [17]. They help to regulate estrogen levels in the body by acting on estrogen receptors in the body. Epidemiological and clinical studies have shown a significant decrease in CRC risk among postmenopausal women who used hormone replacement therapies, providing evidence that CRC is hormone sensitive [18]. In human colorectal tumor cells, estrogen receptor gene expression was diminished or absent; introduction of an exogenous estrogen receptor gene in cultured colon carcinoma cells resulted in marked growth suppression [19]. These studies suggest that endogenous ovarian hormones and even phytoestrogens might modulate CRC risk. Isoflavones also have other chemoprotective properties. One small-scale intervention study showed that soy protein containing isoflavone reduces crypt cell proliferation in colon mucosa biopsies from male and female subjects with a history of colon polyps or colon cancer [20]. Other anticancer activities included a decrease in abnormal cellular proliferation and the induction of apoptosis and inhibition angiogenesis [3, 5]. When examining site-specific differences, currently there are no hypothesized mechanisms for soy contributing to the development of gastric or rectal cancer, but it could be postulated that mechanisms proposed for colon cancer are similar.

Our analysis revealed only a very small decrease in GI cancer risk. This discrepancy may relate to several modifying factors, including the specific type of soy consumed, its preparation and gender differences. Wu et al. [8] recently did a thorough review and a meta-analysis of soy foods and the risk of stomach cancer. Pooled results from the meta-analysis suggested that the risk of stomach cancer may depend on the method of preparation of the soy food. The findings indicated that the risk was lower in association with high intake of non-fermented soy foods and higher with high consumption of fermented soy foods. Messina et al.’s review of in vitro and in vivo data on colon cancer provided similar suggestions [1]. Results from a recent meta-analysis on soy intake and CRC risk, which reviewed 15 case–control and cohort studies, generally suggest an inverse association between higher soy consumption and colon cancer onset; however, nearly all of the confidence intervals overlap 1.0 [7]. Of the six case–control studies that evaluated the association between soy consumption and rectal cancer (732 cases), the point estimates generally suggest an inverse association with unfermented soy consumption and rectal cancer onset but not fermented soy products. The same study also reported a 21 % reduction in CRC risk in women. These findings were not seen in our stratified analyses.

Results from human trials and animal studies have also been conflicting. One human intervention trial providing either a soy or casein supplement for 12 months reported reductions in colon epithelial cell proliferation in the soy group [20]. Another showed that miso dose-dependently lowered the mean numbers of aberrant crypt foci, an early marker for tumorigenesis, in the colons of male rats [21]. On the other hand, of the three studies that specifically examined the effect of soy protein isolate on chemically induced colonic cancer, none found a reduction in the rate of carcinogenesis [10–12]. Furthermore, one study found that soy protein isolate increased colonic cell proliferation [22].

The ability of these studies to detect a true relationship depends on a number of aspects of the assessment of soy food consumption. Results may be influenced by inconsistent measurement tools for evaluating dietary intake, sensitivity of the questionnaire or interview to assess relevant food items for soy intake, and/or the soy content of food. Few studies were originally designed to test the effect of soy as a risk factor or included portion sizes. Measurement standards differed across the studies; the categories of high consumption ranged from a low of one time or more per week to a high of three times or more per day. This inconsistency and the absence of standardized and comprehensive measurements undermine existing evidence. To address this issue, we categorized all relevant studies using standardized measurements of consumption frequency in our analysis to provide a more unified body of evidence.

Potential misclassification can also arise from the difference in composition of soy in the Asian and Western diet that can complicate interpretation of the findings. Asian studies, which comprise the majority of our data, generally measured intake of only one or two types of soy, such as miso and tofu, whereas the soy type was not specified in most studies from Western countries. This leads to potential misclassification and can complicate interpretation of the findings. Results from our stratified analysis did not show any significant difference between studies that examined soy as a whole food group versus specific soy types and method of preparation, suggesting that such modifying factors do not play a prominent role. However, on examining the differences between countries, the correlation of the protective effects of soy with GI cancer was stronger in countries such as China and Japan, as compared with USA. This may relate to higher levels of soy intake and a greater variety on the type of soy consumed.

In addition, as with all questionnaire-based retrospective studies, reliability of results might suffer from measurement and reporting bias. Validation is important to assess the degree to which the questionnaire measures items for which it has been designed. Incorrect information may over or underestimate associations between dietary factors and diseases. Global questions on soy consumption rather underestimate actual intake, whereas a higher number of questions on differently prepared soy foods result in more accurate levels of intake. The observed risk estimate for case–control studies was similar to that for cohort studies, which suggests that in this instance recall bias does not have a prominent effect. Our sensitivity analysis also revealed similar results when non-validated studies were excluded.

Moreover, while most studies adjusted for age and gender in the calculation of risk estimates, not all parameters were considered. Many studies did not appropriately adjust for total energy intake. Total energy intake is mainly a consequence of variations in body size, physical activity and metabolic efficiency. Failure to adjust for such factors can obscure associations between nutrient intakes and disease risk. Intakes of certain nutrients that are correlated with total energy intake may have a non-causal association with disease as a result of confounding by total energy intake. While a meta-analysis would not adequately adjust for this, this is less of a problem for nutrients such as isoflavone that contribute only a small amount of total energy [63].

In summary, the main findings of this study support an inverse association between isoflavone intake and GI cancer risk. Evidence for the chemoprotective effects of soy as a food group in general is much weaker, with only a small decrease in GI cancer risk. The correlation appears to be significant in CRC and among Asian populations. Further research should evaluate isoflavone content within different soy types when measuring exposure, paying attention to the patterns of consumption among different ethnic groups and adjusting for cofounders. Measurements of consumption should be provided in quantifiable terms, using FFQs validated against multiple days of dietary records and adjusting for total energy intake.