Soy Consumption and the Risk of Prostate Cancer: An Updated Systematic Review and Meta-Analysis

Prostate cancer (PCa) is the second most commonly diagnosed cancer in men, accounting for 15% of all cancers in men worldwide. Asian populations consume soy foods as part of a regular diet, which may contribute to the lower PCa incidence observed in these countries. This meta-analysis provides a comprehensive updated analysis that builds on previously published meta-analyses, demonstrating that soy foods and their isoflavones (genistein and daidzein) are associated with a lower risk of prostate carcinogenesis. Thirty articles were included for analysis of the potential impacts of soy food intake, isoflavone intake, and circulating isoflavone levels, on both primary and advanced PCa. Total soy food (p < 0.001), genistein (p = 0.008), daidzein (p = 0.018), and unfermented soy food (p < 0.001) intakes were significantly associated with a reduced risk of PCa. Fermented soy food intake, total isoflavone intake, and circulating isoflavones were not associated with PCa risk. Neither soy food intake nor circulating isoflavones were associated with advanced PCa risk, although very few studies currently exist to examine potential associations. Combined, this evidence from observational studies shows a statistically significant association between soy consumption and decreased PCa risk. Further studies are required to support soy consumption as a prophylactic dietary approach to reduce PCa carcinogenesis.


Introduction
Prostate cancer (PCa) is the second most commonly diagnosed cancer in men worldwide. According to the International Agency for Research on Cancer's GLOBOCAN database, 1.1 million men were diagnosed with PCa in 2012, accounting for 15% of all cancers in men [1]. Incidence rates are lowest in Asian countries, where soy foods are regularly consumed as part of a normal diet. Several studies have reviewed the inverse association seen between soy food intake and PCa incidence in Asian populations, proposing that soy isoflavones act as weak hormones to exert a protective physiological effect against the development of PCa [2][3][4][5]. Indeed, the soy isoflavones, genistein and daidzein, have been shown to accumulate in prostatic tissue [6], where they may be cytotoxic to cancer cells [7]. These effects may occur as a result of both non-hormonal and hormonal action. For example, genistein upregulates tumor suppressor genes in PCa cells [8] and suppresses prostate carcinogenesis in an estrogen receptor (ER) wild-type mouse model, when compared to ER knock-out mouse models [9]. This hypothesis has further been supported by four previous meta-analyses of epidemiological studies, all of which showed a protective association between soy consumption and PCa [10][11][12][13]. However, these meta-analyses did not integrate and evaluate all available existing studies pertaining
When selecting articles for evaluating the association between unfermented or fermented soy food intake and PCa risk, studies had to have explicitly stated which soy food items were being reported. Examples of commonly reported unfermented soy foods included soy milk, tofu and soybeans; fermented soy foods included miso and natto. The pooled RR for unfermented soy foods and risk of PCa was 0.65 (95% CI: 0.56-0.83, p < 0.001), and the pooled RR for fermented soy foods and risk of PCa was 0.86 (95% CI: 0.66-1. 13, p = 0.218) ( Figure 2B,C, respectively). Neither Begg's correlation test (p = 0.161 and p = 0.902, respectively) nor Egger's linear regression test (p = 0.117 and p = 0.670, respectively) for bias were significant. The I 2 index showed high heterogeneity amongst studies included in the unfermented (60.3%) and fermented (66.6%) groups.
Ten articles reported soy intake as a measurement based on the calculation of genistein and daidzein present in soy foods. The pooled RR for genistein and risk of PCa was 0.90 (95% CI: 0.84-0.97, p = 0.008), and the pooled RR for daidzein and risk of PCa was 0.84 (95% CI: 0.73-0.97, p = 0.018) ( Figure 2D,E, respectively). Begg's correlation test was significant for genistein but was not significant for daidzein (p = 0.049 and p = 0.210, respectively), and Egger's linear regression test was significant for both measurements (p = 0.009 and p = 0.039, respectively). The I 2 index showed moderate heterogeneity between studies included in the genistein (31.0%) and daidzein (50.5%) groups. Articles that reported soy intake as either a combination of multiple soy food items or as a single soy food item were classified as total soy intake. Sixteen articles reported the association between total soy intake and PCa risk. The pooled RR for this association was 0.71 (95% CI: 0.58-0.85, p < 0.001) (Figure 2A). Neither Begg's correlation test (p = 0.300) nor Egger's linear regression test (p = 0.052) for bias were significant. Heterogeneity amongst studies was analyzed using the I 2 index to show high variation between studies (68.9%).
When selecting articles for evaluating the association between unfermented or fermented soy food intake and PCa risk, studies had to have explicitly stated which soy food items were being reported. Examples of commonly reported unfermented soy foods included soy milk, tofu and soybeans; fermented soy foods included miso and natto. The pooled RR for unfermented soy foods and risk of PCa was 0.65 (95% CI: 0.56-0.83, p < 0.001), and the pooled RR for fermented soy foods and risk of PCa was 0.86 (95% CI: 0.66-1.13, p = 0.218) ( Figure 2B,C, respectively). Neither Begg's correlation test (p = 0.161 and p = 0.902, respectively) nor Egger's linear regression test (p = 0.117 and p = 0.670, respectively) for bias were significant. The I 2 index showed high heterogeneity amongst studies included in the unfermented (60.3%) and fermented (66.6%) groups. Ten articles reported soy intake as a measurement based on the calculation of genistein and daidzein present in soy foods. The pooled RR for genistein and risk of PCa was 0.90 (95% CI: 0.84-0.97, p = 0.008), and the pooled RR for daidzein and risk of PCa was 0.84 (95% CI: 0.73-0.97, p = 0.018) ( Figure 2D,E, respectively). Begg's correlation test was significant for genistein but was not significant for daidzein (p = 0.049 and p = 0.210, respectively), and Egger's linear regression test was significant for both measurements (p = 0.009 and p = 0.039, respectively). The I 2 index showed moderate heterogeneity between studies included in the genistein (31.0%) and daidzein (50.5%) groups. Six articles reported isoflavone intake without disclosing the sources of the isoflavones. These studies were analyzed separately, so as not to interfere with measurements based solely on soy intake because isoflavones are found in other food items, such as seed sprouts and pulses. The pooled RR for isoflavone intake and PCa risk was 1.03 (95% CI: 0.97-1.09, p = 0.313) ( Figure 2F). Neither Begg's correlation test (p = 0.707) nor Egger's linear regression test (p = 0.802) for bias were significant. The I 2 index showed moderate heterogeneity between studies (44.9%). Notably, the inclusion of these studies in the total soy analysis did not significantly change the RR of dietary soy intake and PCa risk. Forest plots for (A) total soy intake and risk of prostate cancer; (B) unfermented soy intake and risk of prostate cancer; (C) fermented soy intake and risk of prostate cancer; (D) genistein intake and risk of prostate cancer; (E) daidzein intake and risk of prostate cancer; and (F) total isoflavone intake and risk of prostate cancer. These associations were indicated as a relative risk (RR) estimate with the corresponding 95% confidence interval (CI).

Author, Year
Overall (I-squared = 50.5%, p = 0.033) Ward  Forest plots for (A) total soy intake and risk of prostate cancer; (B) unfermented soy intake and risk of prostate cancer; (C) fermented soy intake and risk of prostate cancer; (D) genistein intake and risk of prostate cancer; (E) daidzein intake and risk of prostate cancer; and (F) total isoflavone intake and risk of prostate cancer. These associations were indicated as a relative risk (RR) estimate with the corresponding 95% confidence interval (CI).
Six articles reported isoflavone intake without disclosing the sources of the isoflavones. These studies were analyzed separately, so as not to interfere with measurements based solely on soy intake because isoflavones are found in other food items, such as seed sprouts and pulses. The pooled RR for isoflavone intake and PCa risk was 1.03 (95% CI: 0.97-1.09, p = 0.313) ( Figure 2F). Neither Begg's correlation test (p = 0.707) nor Egger's linear regression test (p = 0.802) for bias were significant. The I 2 index showed moderate heterogeneity between studies (44.9%). Notably, the inclusion of these studies in the total soy analysis did not significantly change the RR of dietary soy intake and PCa risk.
Finally, articles were further stratified into specific soy food groups, which included tofu, miso and soy milk. Few studies were available to accurately represent meta-analysis data on these points, so RR are reported here, but not included in further subgroup analyses. The pooled RR for tofu and PCa risk was 0.73 (95% CI: 0.57-0.94, p = 0.013), the pooled RR for miso and PCa risk was 1.01 (95% CI: 0.80-1.28, p = 0.919), and the pooled RR for soy milk and PCa risk was 0.58 (95% CI: 0.19-1.78, p = 0.343). None of the groups were significant for bias using Begg's correlation test (p = 0.221, p = 0.296, and p = 1.000, respectively) or Egger's linear regression test (p = 0.093, p = 0.497, and p = NA, respectively). The I 2 index showed low heterogeneity amongst the studies included in the tofu (4.5%) and miso (0.0%) groups and high heterogeneity amongst studies included in the soy milk (63.1%) group.

Subgroup Analysis
Articles reporting total soy intake and PCa risk had a pooled RR of 0.61 (95% CI: 0.45-0.82, p = 0.001) for case-control studies and a pooled RR of 0.90 (95% CI: 0.82-0.99, p = 0.022) for cohort and NCC studies. Studies conducted in both North America (p = 0.009) and Europe (p = 0.021) were significantly associated with a reduced PCa risk, whereas studies conducted in Asia (p = 0.064) were not. A complete subgroup analysis can be found in Table 2. Cumulative meta-analyses first demonstrated that soy food intake was significantly associated with the reduced risk of PCa in 1998 and has remained significant over time, and with the inclusion of additional studies within the field ( Figure A3).
Articles reporting unfermented soy food intake and PCa risk had a pooled RR of 0.55 (95% CI: 0.46-0.66, p < 0.001) for case-control studies and a pooled RR of 0.91 (95% CI: 0.76-1.08, p = 0.267) for cohort studies. Studies were conducted in both North America (p = 0.014) and Asia (p = 0.005), and there was a significantly reduced risk of PCa in both continents. Studies of medium quality were significantly associated with a lower PCa risk (p < 0.001).
The pooled RR for articles reporting dietary genistein intake and PCa risk was 0.81 (95% CI: 0.68-0.96, p = 0.016) for case-control studies and a pooled RR of 0.93 (95% CI: 0.85-1.01, p = 0.077) for cohort and NCC studies. Studies conducted in Asia (p = 0.004) showed a significantly reduced risk of PCa, while studies conducted in North America (p = 0.145) and Europe (p = 0.419) did not. Mid quality studies were significantly associated with a reduced risk of PCa (p = 0.007).
The pooled RR for articles reporting dietary daidzein intake and PCa risk was 0.68 (95% CI: 0.47-1.00, p = 0.052) for case-control studies and a pooled RR of 0.91 (95% CI: 0.84-1.00, p = 0.042) for cohort and NCC studies. Studies conducted in Asia (p = 0.012) showed a significantly reduced risk of PCa, while studies conducted in North America (p = 0.094) and Europe (p = 0.799) did not. High quality studies were significantly associated with a reduced risk of PCa (p = 0.042).
There was no significant association for case-control or NCC studies and risk of PCa for articles evaluating circulating levels of genistein. Only studies conducted in Asia (p = 0.031) were significantly associated with a decreased risk of PCa, while studies conducted in Europe (p = 0.784) showed no significant association with circulating genistein levels and PCa risk.
Articles evaluating dietary isoflavones, fermented soy intake, and circulating levels of daidzein showed no significant associations on risk of PCa when studies were grouped by design or by continent. Finally, no studies significantly affected any of the pooled RRs when conducting sensitivity analyses for each group.

Discussion
This updated systematic review and meta-analysis provides a thorough evaluation regarding the association between soy food intake and PCa risk. Using the current pool of scientific literature, our results support the existing evidence, which indicates that total soy food intake is associated with a reduced risk of PCa (p < 0.001). This population-based evidence corroborates observations in both in vitro and in vivo studies, which have shown that soy isoflavones inhibit PCa development and growth [57][58][59]. In agreement with this, we found that both genistein and daidzein intake were inversely associated with the risk of PCa (p = 0.008 and p = 0.018, respectively). These results support our finding that total soy food consumption is associated with decreased PCa risk, as genistein and daidzein are likely found in similar food products.
Soybeans and soy food products contain isoflavones-predominantly genistein and daidzein-mainly as β-glycosides [60]. During digestion, these glycosides are hydrolyzed to their aglycone forms by intestinal or bacterial β-glucosidases [60]. By removing the sugar molecule, the isoflavones are smaller and more hydrophobic, allowing them to more readily diffuse into enterocytes [61]. After absorption and first pass metabolism, aglycones are re-conjugated in the liver to their glycosidic or other conjugated forms and distributed to tissues via systemic circulation [61]. Once within cells, isoflavones act as weak estrogen receptor (ER) agonists or antagonists, depending on the cell type and concentration of estrogen present [61]. Prostatic tissues have higher concentrations of ER-β, to which genistein preferentially binds, with an affinity similar to that of the endogenously-produced estrogen, 17β-estradiol [62]. Increased presence and activation of ER-β is associated with reduced cell proliferation and reduced PCa histological grade [63,64]. This effect has been shown to occur, in part, by reducing the levels of prostate-specific antigen (PSA), cyclin D1, and cyclin-dependent kinase 4 (CDK4) in an ER-dependent manner [57,58]. Interestingly, ER-β expression is often lost during prostate carcinogenesis, so the ability of genistein to bind to ER-β may be a key factor in the inhibition of prostate carcinogenesis. Additional mechanisms of the effects of soy isoflavones on PCa cellular proliferation, apoptosis, and differentiation have been reviewed in depth by Mahmoud et al. (2014) [65].
We also analyzed the potential relationship between unfermented or fermented soy food products and risk of PCa. We found that unfermented soy food products were associated with a decreased risk of PCa (p < 0.001), while fermented soy food products had no associations with PCa risk (p = 0.281). More studies provided food intake data for unfermented soy food products than fermented soy food products (11 studies versus 8 studies, respectively). While this meta-analysis failed to demonstrate a significant association between fermented foods and the risk of PCa, it should be recognized that there was wider variation in results reported by these studies than there was for studies using unfermented and other soy foods. This wider variation could have impacted the risk outcomes. Some concerns have been expressed in the literature regarding the effects of soy fermentation on the risk of developing certain cancers, such as gastric cancer [66]. Due to this association, Yan and Spitznagel chose to not include fermented soy foods in their 2005 meta-analysis [10]. However, during fermentation, β-glucosidases, secreted by fermentative bacteria, cleave glycosidic linkages via a similar process that digestive enzymes in the small intestine and gut microbiota cleave these linkages [60,61,67]. Isoflavones are present in fermented foods, such as tempeh and miso, predominantly as aglycones, with few isoflavones retaining their side-chains. The ratio is reversed for nonfermented foods, but some naturally occurring plant β-glucosidase activity allows for continuous side-chain cleavage to yield aglycones [60]. The more bioavailable aglycone form is readily absorbed from the intestines, rendering this conversion from a glycosylated isoflavone to its aglycone counterpart essential for maximal isoflavone absorption. These similar enzymatic processes yield common aglycone products for intestinal absorption, regardless of whether these processes occur before digestion during the fermentation of soy foods or during the digestion of unfermented soy foods.
Circulating levels of genistein and daidzein were not associated with the risk of PCa (p = 0.236 and p = 0.310, respectively) despite the elevation of circulating genistein and daidzein levels after consumption of these isoflavones. However, peak circulating isoflavone levels can occur as soon as 30 min or as long as 6 h after feeding, depending on the specific isoflavones and foods consumed [61,68]. Mean half-life values vary as well; the average half-life for free genistein is 3.2 h and the average half-life for free daidzein is 4.2 h [69]. Because isoflavones can exist with multiple side-chains (e.g., glucose, sulfate, acetyl, or malonyl-CoA groups) and as different metabolites, accurately measuring in vivo pharmacokinetics is challenging. Thus, the amount of time between isoflavone consumption and blood collection may substantially affect measurement outcomes, which can impact the reliability of blood isoflavone measurements as markers of soy food intake. Due to these factors, an association may have been missed for these circulating isoflavones.
No association was seen when we examined six articles that did not disclose the source of isoflavones measured from participants' dietary intakes. For example, Bosetti et al. (2006) indicated that isoflavones were measured from the FFQ, primarily based on consumption of soy and soy products, but also from "vegetable or bean soups and pulses" [28]. Because isoflavones are found in other food sources and supplements, such as clover and alfalfa seeds and sprouts, garbanzo beans, and other pulses, we independently analyzed studies that did not explicitly indicate that measurements were taken from soy food sources. As such, total dietary isoflavones were not significantly associated with risk of PCa (p = 0.313). While most dietary isoflavones are consumed from soy food products, examining the sources and types of isoflavones that were included in these analyses could provide insight as to why no association was observed. Notably, this analysis was based on the information found from a limited number of articles, so additional studies could strengthen these observations.
To our knowledge, this is the first meta-analysis to investigate the risk associations between soy and advanced PCa. Advanced PCa is defined as poorly differentiated, aggressive, and metastatic disease. Advanced PCa is often difficult to treat, as patients are typically less responsive to therapy. It is therefore imperative to identify other ways to prevent disease progression, such as through dietary modification. Relatively few studies have examined the relationship between diet and advanced PCa or reported information pertaining to stage or grade of PCa. Our results do not show a significant reduction in the risk of advanced PCa with total soy intake (p = 0.119) or dietary and circulating levels of genistein (p = 0.381), daidzein (p = 0.227), or total isoflavones (p = 0.337), perhaps due to the lack of studies. Two double-blinded, randomized, placebo-controlled clinical trials have supplemented isoflavones in men awaiting radical prostatectomy. One study reported higher apoptotic activity in tumors of men treated with isoflavones when compared to tumors of the men in the placebo group [70], while the other study showed modulation of both cell cycle and apoptotic genes in the prostate tissues of men in the treatment group, when compared to tissues of men in the placebo group [71]. More studies are needed to further explore this promising relationship and to identify whether soy can protect against advanced PCa.
Few studies have analyzed individual soy foods and their relationships with PCa risk. Tofu was the most investigated soy food found in the literature. Tofu showed a significant protective association with PCa (p = 0.013). This result is consistent with the result shown by   [11]. More studies are needed to understand the role of individual soy foods in PCa risk.
The research design of studies did not seem to bias our results, as case-control and cohort studies both reported significant and null results. It is important to note that case-control studies are generally considered to have a higher risk of bias than cohort studies; however, significant measurement error can occur for both cohort and case-control studies when evaluating a single exposure variable. To accurately account for all reported estimates of soy exposure and PCa risk, both study designs were included in the analysis. To this end, sensitivity analyses conducted for each subgroup showed that no studies significantly altered results or heterogeneity within subgroups. Separating results by continent showed that total soy intake was only associated with a decreased risk of PCa in North America (p = 0.009) and Europe (p = 0.021), although only one study was included in the European group. However, when looking at the individual dietary isoflavones, genistein and daidzein, Asia was the only continent to show a significant risk reduction for PCa (p = 0.004 and p = 0.012, respectively). Similarly, Asia was the only continent to show a significant association between circulating genistein and risk of PCa (p = 0.031). Both North America (p = 0.014) and Asia (p = 0.005) showed a reduced risk of PCa when only considering unfermented soy foods. The variability in these associations makes it difficult to draw conclusions about whether ethnic differences, preparation methods, or eating patterns exist in soy food or isoflavone consumption and PCa risk.
Because differences in overall dietary composition or other unmeasured lifestyle factors could contribute to increased or decreased disease risk, the results from this study and others showing that soy intake is associated with a reduced risk of PCa should be interpreted with caution. Our study was limited in that our results relied on the reporting of the studies included in this analysis and may have been affected by several factors. For example, studies relying on dietary recall or FFQ reporting are subject to recall bias by participants. Soy food intake measurements could be inconsistently reported or nutrient analyses may differ based on the amount and type of soy food or the database used to collect nutrient information. In addition, not all studies accounted for potential confounding variables, such as family history of prostate cancer (FHPC), body mass index (BMI), smoking, or energy intake. The subgroup analyses in this study attempted to account for these limitations by highlighting some of these differences between studies, to account for variability in data adjustments and selection bias created by study design (i.e., whether the study was case-control or NCC/cohort).
In addition to attempting to address these study limitations, our analysis delved deeper into potential confounding factors, through meta-regression, to determine whether study quality, length, or the sample size impacted significant results of the study. None of these factors were found to impact our results, and as such were not included in our results. We also reported both dietary intakes and circulating levels of isoflavones to create a more comprehensive review of the existing literature. Finally, we analyzed any associations between soy and advanced PCa, which has not previously been reported.
As the second most commonly diagnosed cancer in men worldwide, it is important to identify modifiable factors, such as diet, that may impact the risk of developing PCa. The current study provides an updated systematic review and meta-analysis of the available literature describing the associations between soy food consumption and PCa risk. Of the four meta-analyses previously published, all showed that soy intake was associated with a reduced risk for PCa. Our study further enhances this association by including additional studies for analysis, grouping soy foods by type of food and by isoflavone intake, adding groups of circulating isoflavone concentrations, and by evaluating the potential relationship between soy food intake and advanced PCa risk.        Figure A3. Forest plot for total soy intake and risk of prostate cancer by year of study publication. These associations were indicated as a relative risk (RR) estimate with the corresponding 95% confidence interval (CI).