Flavan-3-ols consumption and cancer risk: a meta-analysis of epidemiologic studies

Although numerous in vitro studies and animal model data have suggested that flavan-3-ols, the most common subclass of flavonoids in the diet, may exert protective effects against cancer, epidemiologic studies have reported inconclusive results for the association between flavan-3-ols intake and cancer risk. Therefore, we conducted this meta-analysis of epidemiologic studies to investigate the preventive effects of flavan-3-ols on various types of cancers. A total of 43 epidemiologic studies, consisting of 25 case-control and 18 prospective cohort studies, were included. A significant inverse association was shown between flavan-3-ols intake and the risk of overall cancer (relative risk (RR) 0.935, 95%CI: 0.891-0.981). When cancer types were separately analyzed, a statistically significant protective effect of flavan-3-ols consumption was observed in rectal cancer (RR 0.838, 95%CI: 0.733-0.958), oropharyngeal and laryngeal cancer (RR 0.759, 95%CI: 0.581-0.993), breast (RR 0.885, 95%CI: 0.790-0.991) in case-control studies and stomach cancer in women (RR 0.633, 95%CI: 0.468-0.858). Our analysis indicates the potential benefits of flavan-3-ols in cancer prevention.

To date, epidemiological studies have focused primarily on flavan-3-ols and cardiovascular disease, and meta-analyses have suggested a reduced risk of cardiovascular disease associated with flavan-3-ols intake [4,5]. Furthermore, when focusing on cancer preventive effects, meta-analyses of observational studies have mainly indicated an association between cancer risk and intake of other flavonoids, such as isoflavones [6,7]. Although numerous in vitro and animal model studies have suggested that flavan-3-ols act as anticarcinogens through antioxidant [8,9] and detoxifying effects [10], modulation of the cell cycle and apoptosis [11][12][13], stimulation of the immune system [14] and DNA repair [15], and suppression of metastasis [16,17] and inflammation [17][18][19], data from human population studies are limited and conflicting [20].

Research Paper
To our knowledge, no quantitative evaluations of the association between flavan-3-ols intake and the risk of overall cancer have been reported. Thus, we performed a quantitative meta-analysis of the currently available epidemiologic studies to estimate the effects of flavan-3ols consumption on cancer risk.

Study characteristics and quality assessment
The main characteristics of the included studies are shown in Table 1. The types of study design were as follows: cohort studies (n=18), hospital-based casecontrol studies (n=18), and population-based case-control studies (n=7). 13 studies were conducted in the Americas (USA, n=11; Mexico, n=1; and Canada, n=1), 29 studies were conducted in Europe (Multiple European countries, n=5; Italy, n=11; Greece, n=4; France, n=1; UK, n=2; Finland, n=2; Netherlands, n=3; and Spain, n=1), and only one study was conducted in Asia (Korea, n=1). Of these studies, 7 evaluated catechins as a major flavan-3ols [33][34][35][36][37][38][39] and others assessed flavan-3-ols according into the USDA databases for the flavonoid content of selected foods [1] (Supplementary Table 1). These studies covered 14 different types of cancer, most of which were adjusted for a wide range of confounders, including age, total energy intake, BMI, physical activity, smoking status, alcohol intake, and total fruit and vegetable consumption. All studies used food-frequency questionnaires to measure flavan-3-ols intake.  The quality of the included studies varied, as assessed using the Newcastle-Ottawa Scale, as shown in Tables 2 and 3. The ranges of quality scores were 8-10 in the cohort studies and 6-9 in the case-control studies. Therefore, all of the cohort studies (n=18) and 84% of the case-control studies (n=21) were high-quality studies (studies with a score≥7).

Overall cancer
Our analysis of the 43 studies showed a 6.5% reduction in the risk of overall cancer with high flavan-3-ols consumption (RR 0.935, 95%CI: 0.891-0.981), as shown in Figure 2. Based on Begg's test, there was no indication of publication bias (P=0.07). However, the asymmetry of Egger's funnel plot suggested a possible absence of negative studies (P=0.003). According to the trim and fill analysis, 12 such studies were filled in to the right of the mean to make the funnel symmetrical, and the adjusted estimated effect was RR 0.987 (95%CI: 0.936-1.040) based on the random-effects model. In sensitivity analyses, each study was excluded one by one to determine its influence on the summary risk estimate. The RR of the remaining studies ranged from 0.928 (95%CI: 0.882-0.977) to 0.952 (95%CI: 0.911-0.994) by omission of Zamora-Ros, 2013a and Cui, 2008 studies, respectively, without great fluctuation. Substantial heterogeneity was observed in this analysis (Q=79.002, P=0.000, I 2 =46.837%), which confirmed that flavan-3-ols have different effects on cancer risk, depending on the cancer site.

Subgroup analyses
Christensen,2012 1 A study could be awarded a maximum of one star for each item, while two stars for the item Control for important factor or additional factor. 2 A maximum of two stars could be awarded for this item. Studies similar to others in terms of patient characteristics (age, sex, BMI) received one star, whereas studies that controlled for other important confounders such as smoking and drinking received an additional star.   [33,103] effect was observed for bladder, breast, colon, esophageal, gastric, liver, lung, ovarian, pancreatic, prostate, renal, or uterine cancer.

DISCUSSION
A total of 43 epidemiologic studies on the effects of flavan-3-ols on cancer risk were identified in this review. To our knowledge, this report is the first quantitative meta-analysis to show an association between flavan-3-ols intake and the risk of overall cancer. Our data suggested that high flavan-3-ols consumption was associated with a 6.5% reduction in the risk of overall cancer. The substantial heterogeneity in the analysis confirmed that the effects of flavan-3-ols differed by cancer site. In particular, flavan-3-ols intake significantly reduced the risk of rectal cancer, oropharyngeal and laryngeal cancer, and thyroid cancer. Reductions in the risk of breast in casecontrol studies and the risk of gastric cancer in women were also observed. However, no effects on the risk of bladder, colon, esophageal, liver, lung, ovarian, pancreatic, prostate, renal, or uterine cancer were observed. In addition, the association between flavan-3-ols intake and cancer risk was observed in Europe but not in Americas or Asia. The possible explanation could be ethnic differences in food structure and genetic backgrounds.
Although this meta-analysis suggested that there was no significant association between colorectal cancer risk and flavan-3-ols consumption, several mechanisms supported by in vitro and animal studies remain biologically plausible. More specifically, in vitro studies have shown that flavan-3-ols inhibit cell proliferation and induce apoptosis in colorectal cancer cells [40][41][42]. In addition, animal studies have shown that catechin, a major flavan-3-ols, is able to decrease the number of intestinal tumors. One meta-analysis found an inverse association between flavan-3-ols consumption and colorectal cancer [43], which was inconsistent with our findings, likely because the analysis did not include the most recent study on the topic [44]. In particular, we found that a significant protective effect of flavan-3-ols was observed in cases of rectal, but not colon cancer. Although colon and rectal cancer are often combined, the differences between colon and rectal cancer in terms of incidence rate [45,46], molecular characterization [47,48], and risk factors [49] may explain the distinct effects of flavan-3-ols on different parts of the intestine.
No effect of flavan-3-ols on the risk of breast cancer was identified in the present study, which is similar to the findings of a meta-analysis by Hui et al [50], although that analysis included one study that evaluated urinary excretion of polyphenol metabolites (which we excluded) [28]. We excluded studies that evaluated urinary or plasma flavan-3-ols because urinary excretion of catechin metabolites has shown a weaker correlation with intake [51] and because plasma values vary substantially throughout the day [52]. When the analysis was stratified by study design, a statistically significant association between breast cancer risk and flavan-3-ols intake was observed in case-control studies only. The inconsistent conclusions between the two different study designs may be attributed to the presence of more recall and selection biases in case-control studies. For example, case subjects may have remembered their diet differently after a diagnosis of cancer compared with control subjects. Therefore, recall bias, with a misclassification of subjects, could potentially have led to a spurious association. In addition, the potential for selection bias should be considered because hospital-based control subjects who suffer from other diseases may tend to change their dietary patterns and because population-based control subjects who agree to participate in a study may be more health conscious than the general population.
Separate analyses by sex showed a significant inverse association between flavan-3-ols intake and the risk of stomach cancer in women. A Japanese pooled analysis of six cohort studies showed similar results, with green tea consumption (>5 cups/day) decreasing the risk of stomach cancer in women, but not in men [53]. These gender differences could be partly explained by the regulation of sex hormones, such as estrogens. Moreover, a large prospective study reported that associations were observed between the risk of stomach cancer and age at menopause, years of fertility and years since menopause, indicating that female hormones play a protective role in stomach cancer risk [54]. Furthermore, the risk of stomach cancer was reduced in women undergoing hormone replacement therapy, and the risk increased after treatment with an antiestrogenic agent [55,56]. Flavan-3-ols such as epigallocatechin gallate (EGCG) and epicatechin gallate (ECG) may inhibit estradiol metabolism and enhance uptake, resulting in a moderate increase in estradiolinduced responses [57]. However, too few publications were included in this subgroup analysis to draw any definite conclusions. Further studies are needed to provide insight into the relationship between sex and flavan-3-ols intake.
Several reviews have reported statistically significant inverse associations between lung cancer risk and total dietary flavonoids intake [58,59]. However, no significant correlations with flavan-3-ols intake have been found [59,60]. Similar to the finding of our metaanalysis, no reduced risk of lung cancer was observed in prior estimates of the effects of flavan-3-ols, including in subgroup analyses by sex, smoking status, and study design. The differences in estimated effects between total dietary flavonoids and flavan-3-ols may have been caused by the contribution of other flavonoid compounds, such as quercetin, kaempferol and isoflavones [7,60] Similar to all meta-analyses, the present study has several limitations. First, the USDA Database for the Flavonoid Content of Selected Foods, which was the first to estimate flavan-3-ols consumption based on dietary data, was published in 2003 [1], so most of the investigations reviewed here were conducted after publication of this flavonoid database. Consequently, limited numbers of previous studies identifying the effects of flavan-3-ols on cancer risk could be included in the current analysis compared with meta-analyses of other flavonoids [7,61]. Second, data from a single food-frequency questionnaire may not accurately reflect flavan-3-ols consumption, which would likely lead to nondifferential misclassification and an underestimation of the true associations. Furthermore, flavan-3-ols content lacks an accurate biomarker and is influenced by a multitude of factors, such as sunlight, season and food processing [1,51], thus contributing to inconsistency between studies. Third, because only published studies were included in the present analysis, the analysis may have been affected by publication bias. In fact, based on Egger's test, publication bias was indicated in the analysis of overall cancer. However, the sensitivity analysis showed no change in statistically significant levels, suggesting that the observed potential bias may not have substantially affected the summary measures. Finally, substantial heterogeneity was observed in the analysis of overall cancer, which was not surprising because of the variation in the effects of flavan-3-ols on cancer risk by cancer site. Therefore, subgrouping by cancer site reduced the apparent levels of heterogeneity.
The result from random-effect linearity metaregression model suggested that there was no doseresponse relationship between flavan-3-ols and cancer risk (data not shown). Furthermore, there are no data to suggest that long term use of high dosages of flavan-3-ols is detrimental to health. However, experimental studies based on physiologic dose levels are still needed to clarify an adequate range of exposures of flavan-3-ols.
To our knowledge, this is the first meta-analysis to comprehensively assess the effects of flavan-3-ols on the risk of a range of cancers. In the future, the findings should be confirmed by more well-designed cohort or intervention studies. And the underlying mechanisms of the different preventive effects of flavan-3-ols in various cancers remain to be elucidated.

Search strategy
The PubMed, EMBASE and Cochrane databases were searched from their inception to January 2016 using the following terms in combination: flavan-3-ols (catechin, flavonoid), cancer (neoplasm, tumor, carcinoma) and epidemiology (case-control, cohort). In addition, the references of the retrieved articles were reviewed, and the searches were not limited by language.

Study selection
An article was included in our meta-analysis if it (1) described a cohort or case-control study, (2) assessed flavan-3-ols (catechin) consumption via food intake, (3) provided a risk estimate (hazard ratio (HR), relative risk (RR), or odds ratio (OR)) and 95% CIs for cancer incidence, and (4) provided information on adjustment for confounding factors. Studies that evaluated urinary or plasma flavan-3-ols were excluded from this analysis. When an article could not be rejected with certainty based on the title and abstract, the full text was obtained, and inclusion was assessed independently by 2 reviewers (LL and YY).

Data extraction and quality assessment
Data were extracted independently by 2 authors (LL and YY) using a common form, and differences were subsequently adjudicated. The standardized data extraction form collected the following: the study name (the first author's last name and the year of publication); the region; the study design; the type of controls for case-control studies (population-based or hospital-based controls); the sample size (cases and controls or cohort size); the follow-up time for cohort studies; the type of flavonoid intake; the comparison of flavan-3-ols intake; the type of cancer; the adjusted RR, OR or HR with 95% CIs for the highest compared with the lowest amount of flavan-3-ols consumed; and the most completely adjustment.
The assessment of study quality used the Newcastle-Ottawa Scale, which is a 9-star system including three dimensions: selection; comparability; and outcome (cohort studies) or exposure (case-control studies), depending on the study type [62]. An additional point was awarded when a study analyzed an energy-adjusted residual or nutrientdensity model [7,63], considering that the present metaanalysis assessed nutrient consumption and cancer risk. A study with≥7 points was considered to be a high-quality study using this 10-point system.

Statistical methods
As the incidence of cancer was relatively low over the period of time studied, the most completely adjusted OR or HR (highest versus lowest flavan-3-ols intake) was considered to be an approximation of the RR. Studies were combined using the random-effects model, in which the effect measures are logRR weighted using the method of DerSimonian and Laird [64]. Statistical heterogeneity among studies was estimated using Q and I 2 statistics [65]. To assess publication bias, funnel plots were constructed, and the methods of Begg's test (rank correlation) [66] and Egger's test (linear regression) [67] were used to test funnel plot asymmetry. If a potential bias was detected (P < 0.05), the trim and fill analysis was further used to examine the impact of possibly missing studies [68]. To assess whether the combined estimates could have been markedly affected by a single study, a sensitivity analysis was also conducted, in which each study was excluded one by one and the analysis was repeated based on the remaining studies. All data analyses were performed with Comprehensive Meta-Analysis Software, version v. 2.0 (CMA, Biostat, Englewood, NJ, USA).