Background

Tea is a popular beverage consumed worldwide, generally in the forms of black and green tea. Tea is produced from the leaves of the Camellia sinensis plant through several processes. Black tea is the main tea beverage in the United States, Europe, and Western Asia, while green tea is more popular in China, Japan, and Korea [1]. Extensive laboratory studies using multiple animal models have suggested that tea and tea polyphenols might have an inverse association with cancer through its apoptosis-inducing, anti-mutagenic, and antioxidant properties [2, 3].

In some recent reviews, researchers have suggested that green tea, which contains abundant polyphenols and catechins, specifically epigallocatechin-3-gallate (EGCG) 5, might have a protective effect against cancers. Using multiple approaches, studies have shown that polyphenols, theaflavins (TF) and thearubigins (TR) in black tea might possess chemopreventive properties. However, most of the evidence showing a protective effect of tea on cancer has been generated in animal experiments but has not been demonstrated in human trials [4, 5].

The World Cancer Research Fund report of 2007 concluded that the evidence for associations between the consumption of tea and risk of some major cancers was still limited and inconsistent [6]. The results from a few clinical trials and epidemiological studies also indicated that the preventive effect of tea or its extract on cancer is controversial. In a recent clinical trial evaluating the efficacy of green tea extract (GTE) on prostate cancer, it was found that the GTE had minimal clinical activity [7, 8]. However, another Phase II clinical trial suggested that higher doses of GTE might improve the short-term outcome in patients with a higher risk of oral premalignant lesions [9]. In a cohort study conducted in the USA, tea consumption was found to have no inverse association with colorectal cancer, and the hazard ratio (HR) changed only slightly as tea consumption increased [10]. However, in another cohort study conducted in China, results showed that regular green tea consumption was associated with a reduced risk of colorectal cancer in smokers [11]. Some recent systematic reviews have revealed conflicting results between meta-analyses with prospective studies and those with retrospective studies [1216].

Previous meta-analyses mainly focused on the relationship between the highest tea consumption level and either the lowest tea consumption level or non-drinkers. However, the range of tea consumption and the cut-offs for the categories differed between studies. Another drawback of the previous meta-analyses is the inclusion of retrospective case–control studies, which were sensitive to confounding variables and bias, especially recall bias. To quantitatively assess the relationship between tea consumption and risk of five major cancers, we conducted a systematic review and dose–response meta-analysis with prospective studies. The five major cancers we studied were liver, stomach, breast, prostate, and colorectal cancer.

Methods

Literature search

We performed a systematic literature search in PubMed, Embase, and Cochrane Library with a combination of the following terms: tea AND (breast OR prostate OR stomach OR gastric OR colorectal OR colorectum OR rectal OR rectum OR colon OR large bowel OR liver OR hepatic OR hepatoma) AND (cancer OR cancers OR carcinoma OR carcinomas OR neoplasm OR neoplasms). No language restrictions were imposed. Reference lists of the identified publications were also reviewed for inclusion/exclusion. We also searched the conference abstract on the website of American Society of Clinical Oncology (ASCO) annual meeting from 2004 to 2013. Two reviewers (FY and CX) independently selected studies based on the titles and abstracts of the retrieved studies. Studies were included if they met the following criteria: (1) a prospective study assessing the association between tea consumption and at least one of the five selected cancers (breast, stomach, colorectal, liver and prostate cancer); (2) a study considering at least three levels of tea consumption and providing a sample size for cases and non-cases in each exposure category [17]; and (3) a study reporting the relative risks (RRs) of different dose categories with 95% confidence intervals (95% CIs) adjusted for sex, age, or other factors. We excluded retrospective case–control studies because of their inherent limitations, especially recall bias. However, nested case–control studies and case-cohort studies were included in our meta-analysis because the at-risk study populations in each of the exposure categories are derived from cohort studies and diet exposure would have been investigated years before the onset of cancer, which would technically eliminate recall bias. We also excluded studies that reported total tea consumption on monthly, weekly, or daily basis, but did not provide data on number of cups or times per month, week or day. One study assessing breast cancer risk was excluded because we could not extract data stratified by sex in both exposed and unexposed groups [18]. Another study was excluded because it reported stomach cancer risk based on iced tea and hot tea consumption, but no other study provided such data [19].

Data extraction

Data extraction was performed according to the MOOSE (meta-analysis of observation studies in epidemiology) guidelines [20] (see the Additional file 1) by two reviewers (FY and CX), and verified independently for accuracy by a third reviewer (ZJ). Discrepancies were discussed with a fourth reviewer (JH) to reach a consensus. For each included study, the following data was extracted: the title and author of the study, publication year, study population, study location, sample size, proportion of males, median of follow-up time, the type of consumed tea, covariates controlled for by matching or multivariable analysis, the numbers of cases/non-cases, total person-years, relative risk (RR) of the different exposure categories and the corresponding 95% confidence intervals (95% CIs), response rate, and how exposure were assessed. For studies that reported several multivariable adjusted RRs, we selected the effect estimate that adjusted for the maximum potential confounders. The quality of the included studies was assessed according to the 9-star Newcastle-Ottawa Scale (NOS) [21] by two investigators (FY and HJ).

Statistical analysis

We performed a two-stage dose–response meta-analysis to examine the relationship between tea consumption and five major cancer risks. First, we synthesized the RRs across categories of tea consumption in each study [22, 23]. Because the absolute risk of cancer is low, the odds ratios (ORs) in nested case–control studies approximated the RRs [24]. Pooling of RRs from each study requires the exposure levels and distribution of cases and person-years or non-cases in each category of tea consumption. However, not all studies reported the distributions of cases and person-years or non-cases for exposure categories. Nine studies did not report individuals or person-years for each category and instead reported the total sample size [10, 11, 2531]. We estimated the distribution of cases for each category in such studies by using the methods described by Aune [32]. To assess exposure levels, we converted all measures into cups per day and defined 125 mL of tea as one cup regardless of tea type unless it was well established in a specific study population or a geographical area. If the study reported tea consumption as number of times, we regarded one time as one cup. As some Chinese studies reported the amount of tea leaves consumed as the measure of tea consumption, we regarded consumption of 150 g of tea leaves per month as one cup per day; this allowed us to universalize all of the included studies in a single standard unit [33].

If a study did not report the median of the exposure category, we assigned the level of tea consumption to categories based on the calculated midpoint of tea consumption. When the highest category was open-ended, we assumed the dose as 1.2 times the lowest bound of this category [17]. In studies reporting tea consumption by cups per month or cups per week, we redefined these exposure categories as cups per day by multiplying with 1/30 or 1/7, respectively. Subsequently, we estimated the overall RR by combining the RRs derived from the first step. A fixed effect model was used if there was no evidence of heterogeneity; otherwise a random effect model was adopted [34, 35]. Forest plots were used to visually assess the RR estimates and corresponding 95% CIs. We also tested the nonlinear relationship between tea consumption and cancer risk by modeling tea consumption levels by using restricted cubic splines with 3 knots at fixed percentiles (10%, 50%, and 90%) of the distribution as described by Larsson and Orsini [36, 37]. A P value for nonlinearity was calculated by testing the null hypothesis that the coefficient of the second and third spline was equal to zero.

Eight studies on colorectal cancer [10, 26, 27, 31, 3841] and three studies on stomach cancer [27, 42, 43] reported cancer risk by sex. Ten studies reported colorectal cancer risk stratified as colon or rectal cancer [10, 18, 26, 27, 3841, 44, 45]. One study reported colon cancer risk as a distal or proximal cancer [38]. The study by Inoue et al. pooled data from six cohort Japanese studies that studied the relationship between tea consumption and stomach cancer [43]. Results from these studies were first pooled by using a fixed model and then included in the overall risk estimate. The χ2 test and I2 statistic were used to explore the heterogeneity among studies [46]. The Egger’s regression test, Begg’s rank correlation test, and visual inspection of a funnel plot were performed to assess publication bias [47, 48]. As a rule of thumb, tests for asymmetry should be used only when there are at least 10 studies included in a meta-analysis [35]. We conducted subgroup analyses stratified by sex, tea types, and geographic regions. For breast cancer, we also performed a subgroup analysis stratified by menopausal status. We performed a sensitivity analysis in which one study at a time was removed and the rest analyzed to evaluate whether the results could have been affected markedly by a single study. To detect whether different assessment ways may bias the results, further subgroup analyses were performed by excluding studies that reported tea consumption by frequency (times/servings) [18, 42, 4952], weight of tea leaves (grams) [11, 27, 33], or volume (mL) [53, 54], rather than cups.

We used Stata (Version 12.0; Stata Corp, College Station, TX) for all analyses and all statistical tests were two-sided. P < 0.05 was considered statistically significant.

Result

Study characteristics

As of December 28, 2013, 1,881 records were retrieved by using our search strategy. After reviewing the titles and abstracts, we excluded 1,668 articles and 213 articles were further evaluated by reviewing the full texts. Finally, we identified 41 articles assessing tea consumption and cancer risk, which satisfied the inclusion criteria for our meta-analysis. A flow diagram of study selection is provided as Figure 1. Among the 41 articles, 15 assessed the relationship between tea drinking and the incidence for breast cancer [25, 33, 41, 49, 51, 5463], 15 for colorectal cancer [10, 11, 18, 2629, 31, 3841, 44, 45],[59], 4 for liver cancer [18, 27, 64, 65], 7 for prostate cancer [30, 50, 53, 6669], and 5 for stomach cancer [27, 42, 43, 52, 59]. The 41 articles included had 3,027,702 participants and 49,103 cancer cases. The cancer cases included 20,500 breast cancer patients; 16,202 colorectal cancer patients; 882 liver cancer patients; 4,698 prostate cancer patients; and 6,821 stomach cancer patients. Most of the included studies awarded more than 7 stars according to the Newcastle-Ottawa Scale and were identified as high quality. The general characteristics of the included studies are presented in Table 1.

Figure 1
figure 1

Summary of article selection process.

Table 1 Main characteristics of the studies on tea consumption and five selected cancer included in the meta-analysis

Tea consumption and cancers

The associations between tea consumption and the risk of major cancers are shown in Figures 2, 3, 4, 5 and 6. For breast cancer, the overall RR for three cups increment per day of tea consumption was 1.02 (95% CI, 0.98 to 1.05) with mild heterogeneity among studies (P = 0.22, I 2 = 21.2%). For colorectal cancer, the pooled RR for three cups increment per day was 0.98 (95% CI, 0.93 to 1.03) with mild heterogeneity (P = 0.29, I 2 = 15.0%). For liver cancer, the overall RR for three cups increment per day was 0.91 (95% CI, 0.74-1.12) with moderate heterogeneity (P = 0.10, I 2 = 52.5%). For prostate cancer, the overall RR for three cups increment per day was 1.02 (95% CI, 0.96 to 1.09) with moderate heterogeneity (P = 0.14, I 2 = 37.8%). For stomach cancer, the overall RR for three cups increment per day was 0.98 (95% CI, 0.93-1.03) with moderate heterogeneity (P = 0.15, I 2 = 40.6%).

Figure 2
figure 2

Relative risk estimates of breast cancer per 3 cups increase in tea consumption.

Figure 3
figure 3

Relative risk estimates of colorectal cancer per 3 cups increase in tea consumption.

Figure 4
figure 4

Relative risk estimates of liver cancer per 3 cups increase in tea consumption.

Figure 5
figure 5

Relative risk estimates of prostate cancer per 3 cups increase in tea consumption.

Figure 6
figure 6

Relative risk estimates of stomach cancer per 3 cups increase in tea consumption.

As shown by Figure 7 and the P-value for nonlinearity, we found no evidence of nonlinear relationships between tea consumption and risk of cancers.

Figure 7
figure 7

Dose–response relations between tea consumption and relative risks of breast, colorectal, stomach, prostate cancer.

In the subgroup analysis, we pooled the studies into groups by sex, tea type, geographic region, and menopausal status for breast cancer. The results are shown in Table 2. We found that three cups of black tea consumption increment per day may be a risk factor for breast cancer (RR, 1.18; 95% CI, 1.05-1.32). The result of the subgroup analysis of stomach cancer indicated that tea consumption was a preventive factor (RR, 0.88; 95% CI, 0.80-0.98) in women. However, only three studies were included in this subgroup.

Table 2 Subgroup analysis of cancer risk for an increment of three cup tea consumption by gender, tea type and geographic region

A sensitivity analysis omitting one study at a time and calculating the pooled RRs for the remainder of the studies suggested that no single study dramatically influenced the pooled RRs (results are not shown). After removing the studies that did not report tea consumption as cups per day, the results did not change significantly. On excluding a study by Inoue et al. [43], which had a significantly larger sample size in comparison with other included studies, no significant differences were observed.

Egger’s regression test and Begg’s rank correlation test showed no significant asymmetry of the funnel plot for breast (P = 0.59 and P = 0.60, respectively) and colorectal cancer (P = 0.59 and P = 0.73, respectively), indicating no evidence of substantial publication bias (Figure 8). For the other three types of cancer, we did not perform an analysis for publication bias because of limited numbers of included studies (no more than ten).

Figure 8
figure 8

Funnel plot of log relative risk vs standard error of log relative risks.

Discussion

The findings from our meta-analysis reveal no appreciable association between tea consumption and the relative risk of liver, stomach, breast, prostate, or colorectal cancers. The risk differences were all near zero for the five major cancers with an increase in tea consumption of three cups per day (approximately 375 mL per day). Subgroup analyses, stratified by sex, geographic regions, and type of tea, showed no significant associations with risk of cancer.

The results of heterogeneity test suggested that most pooled effects had mild heterogeneity. However, moderate heterogeneity was seen in studies on liver cancer. Sensitivity analysis performed by omitting one study at a time, identified the main source of heterogeneity as the study by Inoue et al. [64], which included hepatitis C and B virus positive patients while the other three studies did not. After excluding this study, the overall RR was 0.86 (95% CI, 0.75-0.99) with a mild heterogeneity (P = 0.30, I 2 = 17.1%), which suggests a preventive effect of tea drinking for liver cancer. However, large prospective studies are still needed to verify this preventive effect.

Eleven previous meta-analyses have evaluated the association between green tea or black tea intake and the risk of the five selected cancers [1216, 7075]. Of these earlier meta-analyses, three focused on breast cancer, three on colorectal cancer, three on stomach cancer, one on primary liver cancer, and one on prostate cancer. All of the previous eleven meta-analyses reported the summarized RR/OR for the highest tea consumption level relative to the no tea/lowest level. In nine of the meta-analyses that examine the association between green tea consumption and cancer risk, three studies presented an overall statistically significant inverse association between green tea consumption and cancer risk when they included both cohort and case–control studies in the meta-analysis. The summarized ORs of these three studies were 0.82 (95% CI, 0.70-0.96) and 0.86 (95% CI, 0.74-1.00) for stomach cancer and 0.82 (95% CI, 0.69 -0.98) for colorectal cancer [71, 72, 75]. Among the eleven meta-analyses, seven studies reported an inverse association between green tea consumption and cancer risk from the summarized results of case–control studies. However, none of the studies indicated a statistically significant green tea-cancer risk association from a meta-analysis of prospective cohort studies only. These earlier results from prospective studies were consistent with our present meta-analysis. For a concise comparison, we have presented the pooled RRs of previous studies and our study in Table 3.

Table 3 omparison of findings of the present doseresponse meta-analysis with those reported in previous meta-analysis

Among the three meta-analyses examing the relationship between black tea consumption and cancer risk, Sun et al. presented conflicting results across cohort studies and case–control studies for breast cancer risk [73]. The pooled RR for prospective cohort studies was 1.15 (95% CI, 1.02 to 1.31), while the pooled RR for case–control studies was 0.91 (95% CI, 0.84 to 0.98). Our meta-analysis also found a positive association between black tea consumption and the risk of breast cancer in a subgroup analysis. The summarized RR was 1.18 (95% CI, 1.05-1.32) with an increment of three cups black tea intake per day. Association of black tea consumption with breast cancer risk is biologically plausible. Black tea intake has been positively associated the estrogen levels, and experimental studies have established that estrogen is a strong promoter of mammary carcinogenesis [76]. Wu et al. reported that the levels of circulating estrogens were higher in black tea drinkers than in non-tea drinkers [77], and Larsson et al. reported that black tea consumption was positively associated with the risk of ER+/PR + breast cancer, suggesting a possible carcinogenic role involving sex hormones [61].

Of the three meta-analyses exploring the association between tea consumption and colorectal cancer risk, Zhang et al. [74] found a positive association between tea and colorectal cancer (RR 1.28; 95% CI, 1.02-1.61) for tea consumption greater than four cups. Our pooled RR was 0.98 (95% CI, 0.93-1.03) with an increment of three cups black tea intake per day. These differing results could be attributed to the calculation using different increment cups per day. We concluded that drinking three cups more per day does not increase the risk of colorectal cancer while Zhang et al. concluded that drinking four cups increment per day might be a risk factor for colon cancer. In addition, Zhang et al. also reported that the risk of colon cancer did not increase with a 250 g/day increment of tea consumption either in men (RR, 1.03; 95% CI, 0.96-1.09) or women (RR, 1.03; 95% CI, 0.99-1.08); this is consistent with our current meta-analysis. Zhang et al. attributed the observed positive association to chance.

It is notable that most epidemiological evidence for the inverse association between tea intake and cancer risk has come from retrospective case–control studies. Because information in case–control studies is collected after cancer is diagnosed, it reflects the past exposure history based on recall; therefore, recall bias is inevitable and cannot be ignored. This bias may partly explain the difference in the findings between prospective studies and retrospective case–control studies. Given the limitations of case–control studies, the conclusion that tea consumption has an inverse association with cancer risk is not convincing.

When compared with previous meta-analyses, our meta-analysis has several strengths. First, we assessed the association of tea consumption with five major cancers. Second, only prospective studies were included in the meta-analysis, which greatly reduces the likelihood of selection and recall biases. Third, the dose–response analysis included a wide range of tea consumption, which allows a concrete and quantitative assessment of the dose–response relationship between tea intake and cancer risk.

However, several potential limitations of our meta-analysis must be considered when interpreting the results. First, the temperature of tea is an important confounding factor when assessing the association between tea drinking and cancer risk. However, we could not analyze the effect of this factor because only one study reported the relationship between iced or hot tea and cancer; this study reported no significant association were observed between iced or hot tea and gastric cancers. Second, the methods of data collection differed across the included studies. Most of the studies assessed the exposure of the tea consumption via food frequency questionnaire, and the response rate varied among these studies. As is known, people in different countries have different traditions of tea consumption or drinking style, which presents difficulties in assessing tea consumption amounts accurately. In addition, people from the same country may also have different habits of drinking tea, such as strong or weak tea. In addition, tea consumption level is mostly assessed as the number of cups of tea consumed daily or weekly. However, cup size may vary considerably for different countries or areas and the dry tea leaves brewed in each cup may also be different. The cut-offs for the highest consumption level varied across different studies. Therefore, there might be some inevitable measurement errors and possible uncontrolled confounding factors when assessing tea consumption; this could prevent the detection of a modest association between tea consumption and cancer risk. However, we endeavoured to decrease these errors by unifying the unit of measurement and using a similar standard to calculate daily tea consumption. We also performed sensitivity analyses by removing the studies that did not report tea consumption as cups and performed subgroup analyses by geographic region, especially China and Japan (see Table 2). Third, we had no information about the family history of certain cancers in the primary aggregate results. Genetic factors play an important role in the development of cancers. Some genetic factors may increase the susceptibility to cancer, and certain polymorphisms in genes that are responsible for metabolising tea may have a role in the tea-cancer association. Fourth, a meta-analysis is unable to account for confounding factors inherent in the original studies. Although major potential confounders, including age, sex, alcohol, and smoking had been adjusted in most included studies, residual or unknown confounding cannot be excluded as a potential explanation for the observed findings. It is known that alcohol and smoking are important potential confounding factors. For example, people in China who drink tea frequently are more likely to drink more alcohol or smoke. The interplay of these factors, tea, alcohol, and smoking, could not be detected because of limited data. Most of the included studies did not provide the numbers for case and non-case population at each level of tea consumption; therefore, based on the available data, analyses among non-smokers and non-alcohol drinkers could not be conducted. Additionally, not all studies adjusted for these confounding factors. For breast cancer, the menopausal status of women was an important confounding factor. However, according to our subgroup analysis on pre-menopausal (RR, 0.96; 95% CI, 0.79-1.18) and post-menopausal (RR, 1.12; 95% CI, 0.96-1.30) women, there were no significant risk associations between tea consumption and pre-menopausal or post-menopausal breast cancer. Considering only three studies were available for this subgroup analysis, the results may be because of limited sample size. Thus, menopausal status could be another effect modifier for breast cancer, and further studies are needed to discuss the confounding effect of this factor for breast cancer [33]. Finally, publication bias is an inevitable problem in systematic reviews and meta-analyses. However, the results of publication bias analysis showed that there was no significant evidence of publication bias in studies on breast and colorectal cancer. The limited number of included studies precludes us from conducting publication bias analyses for liver, prostate, or stomach cancer.

Conclusion

There is insufficient information from epidemiologic studies to support the suggestion that tea intake plays a role in the prevention of cancer. Randomized controlled trials and large prospective cohort studies are needed to further explore this association.

Author’s contributions

FY developed inclusion and exclusion criteria, drafted and revised the manuscript, contributed to the study design, analysis, and interpretation of the data. ZJ drafted the manuscript and contributed to the study design. HJ contributed to revising the manuscript, data analysis and interpretation of the data. CX developed inclusion and exclusion criteria, contributed to the data analysis and interpretation of the data. JT and TL contributed to the electronic search, hand-search of literature and the revision of the manuscript. JH contributed to the interpretation of the results and the conception of the study. All authors read and approved the final manuscript.