This study was conducted and reported following the preferred reporting items for systematic reviews and meta-analyses (PRISMA) statement (Moher et al., 2009). The inclusion criteria were as following: (1) case-control study or cohort study; (2) exposure was tea (including green tea, black tea, mate, and oolong tea) consumption; (3) outcome was incidence of bladder cancer; (4) study provided the odds ratios (ORs) or relative risks (RRs) with corresponding 95% confidence intervals (CIs) or data necessary to calculate them. When multiple papers reported on the same study were identified, the most informative or complete article would be included.

PubMed and Embase were searched for studies examining the relationship between tea consumption and bladder cancer that were published in any language up to March, 2016. Search items including “tea,” “drink,” “beverage,” or “fluid” combined with “bladder cancer,” “bladder neoplasm,” “bladder tumor,” “bladder carcinoma,” “urothelium carcinoma,” or “transitional cell carcinoma” had been allied in the database retrieve. We also scanned the reference lists from all retrieved papers to identify additional studies. No restriction was applied.

All data were extracted independently and crosschecked by two reviewers according to the pre-specified inclusion criteria. Discrepancy was resolved by discussion. The following information were extracted: First author, publication year, country, study period, sex, study design, type of control subjects for case-control studies, sample size, type of tea, consumption categories, the OR or RR with 95% CI for each category (the results adjusted with most potential confounders), and adjusted variables. Crude ORs or RRs with 95% CIs were only extracted when no adjusted ORs or RRs were presented. In addition, ORs or RRs with 95% CIs in different smoking status were also extracted for assessing the effect of smoking, which is an important confounding factor for bladder cancer.

We performed methodological quality assessment of the included studies using the Newcastle-Ottawa Scale (Stang, 2010; Zeng X. et al., 2015), which is a nine-star scale contained three main items: Selection (0–4 stars), comparability (0–2 stars), and exposure (for cohort study, 0–3 stars), or outcome (for case-control study, 0–3 stars).

The statistical analysis for the overall relationship between tea consumption and bladder cancer risk were based on random-effects model and on comparisons of the highest vs. lowest category of tea consumption (Zeng X. T. et al., 2015). The measure of interest is the OR with corresponding 95% CI. For studies reported the information by subsets (sex, smoking status, type of tea), we summarized the ORs with 95% CIs of the subsets in a fixed-effect model before aggregating them into overall analysis.

For dose-response analysis, we used the G-L method (Greenland, 1987; Orsini et al., 2012) to explore the relationship between tea consumption and bladder cancer risk. The potential non-linearity association was examined by modeling tea intake using restricted cubic splines with three knots at 10, 50, and 95% of the distribution. We assigned the median or middle point of the upper and lower boundaries in each category as the corresponding dose to the related OR for each study. If the highest category is open-ended, we assumed the both boundaries to be the same as the closest category. The lowest boundary was assumed to be zero if it was not present. Studies only reported three levels or more were included in the dose-response analysis.

The Cochran Q and I² statistics were applied to detect statistical heterogeneity among studies (Higgins and Thompson, 2002). Heterogeneity was confirmed with a P-value of less than 0.1 or I² value of more than 50%. To explore the potential heterogeneity among studies, stratified analyses were performed according to study design, sex, study location, smoking status, adjustment for age, and adjustment for smoking. In addition, subgroup analysis of type of tea had been carried out to further investigate the association between different type of tea consumption and the risk of bladder cancer. The meta-regression analysis was conducted to detect the between-group heterogeneity based on the aforementioned variables. Sensitivity analysis was performed by removing the studies that only provided crude ORs to examine the influence of these studies on the summarized Ors (Leng et al., 2015). Publication bias was detected by funnel plot and Egger's regression method (Egger et al., 1997). All statistical analysis was performed with Stata 12.0 (StataCorp, College Station, TX). All statistical tests were two-sided, with P < 0.05 considered statistically significant.

Detailed literature selection process was presented in Figure 1. A total of seven prospective cohort studies (Heilbrun et al., 1986; Chyou et al., 1993; Michaud et al., 1999; Nagano et al., 2000; Zeegers et al., 2001a; Kurahashi et al., 2009; Ros et al., 2011) included 443 076 participants, in which 1807 developed bladder cancer, and 25 case-control studies (Morgan and Jain, 1974; Howe et al., 1980; Hartge et al., 1983; Ohno et al., 1985; Jensen et al., 1986; Risch et al., 1988; Slattery et al., 1988; Clavel and Cordier, 1991; Nomura et al., 1991; D'Avanzo et al., 1992; Kunze et al., 1992; La Vecchia et al., 1992; Wilkens et al., 1996; Bruemmer et al., 1997; Lu et al., 1999; Bianchi et al., 2000; Geoffroy-Perez and Cordier, 2001; Woolcott et al., 2002; Wakai et al., 2004; Bates et al., 2007; De Stefani et al., 2007; Demirel et al., 2008; Jiang et al., 2008; Hemelt et al., 2010; Wang J. et al., 2013) including 15 643 cases and 30 795 controls, published from 1974 through 2013, were identified in the meta-analysis. Characteristics of included studies were presented in Table 1. Of the 32 studies, seven were cohort studies, 12 were hospital-based case-control studies, and 13 were population-based case-control studies. Of these studies, 11 conducted in US (Hartge et al., 1983; Heilbrun et al., 1986; Slattery et al., 1988; Nomura et al., 1991; Chyou et al., 1993; Wilkens et al., 1996; Bruemmer et al., 1997; Michaud et al., 1999; Bianchi et al., 2000; Jiang et al., 2008; Wang J. et al., 2013), 9 in Europe (Jensen et al., 1986; Clavel and Cordier, 1991; D'Avanzo et al., 1992; Kunze et al., 1992; La Vecchia et al., 1992; Geoffroy-Perez and Cordier, 2001; Zeegers et al., 2001a; Demirel et al., 2008; Ros et al., 2011), 4 in Canada (Morgan and Jain, 1974; Howe et al., 1980; Risch et al., 1988; Woolcott et al., 2002), 4 in Japan (Ohno et al., 1985; Nagano et al., 2000; Wakai et al., 2004; Kurahashi et al., 2009), two in China (Lu et al., 1999; Hemelt et al., 2010), one in Argentina (Bates et al., 2007) and one in Uruguay (De Stefani et al., 2007). The estimated quality of all included studies was in the rage of 4–8 stars.

The meta-analysis of all 32 studies, no significant association was observed between high tea consumption and risk of bladder cancer (highest vs. lowest: OR = 0.96, 95% CI 0.86–1.06) (Figure 2), with moderate to high heterogeneity (I² = 54.2%, P_{heterogeneity}). For cohort studies, the pooled OR was 0.88 (95% CI 0.67–1.17), with certain between-study heterogeneity (I² = 67.8%, P_{heterogeneity} = 0.005) (Table 2). For hospital-based case-control studies, the pooled OR was 0.98 (95% CI 0.78–1.24), with certain evidence of between-study heterogeneity (I² = 63.2%, P_{heterogeneity}). For population-based case-control studies, the pooled OR was 1.00 (95% CI 0.91–1.10), with no evidence of between-study heterogeneity (I² = 14.0%, P_{heterogeneity}).

Four cohort studies (Michaud et al., 1999; Nagano et al., 2000; Zeegers et al., 2001a; Kurahashi et al., 2009) and 12 case-control studies (Morgan and Jain, 1974; Hartge et al., 1983; Jensen et al., 1986; Clavel and Cordier, 1991; Kunze et al., 1992; Lu et al., 1999; Bianchi et al., 2000; Woolcott et al., 2002; Wakai et al., 2004; Jiang et al., 2008; Hemelt et al., 2010; Wang J. et al., 2013) were included for the dose-response meta-analysis of tea consumption. The results had been shown in Figure 3. There was no evidence of a non-linearity association between tea consumption and bladder cancer risk (P_{non-linearity} = 0.51). Thus, a linear regression model was applied. The summary OR of bladder cancer for an increase of one cup of tea per day was 1.01 (95% CI 0.97–1.05, P_linear = 0.73). For cohort studies, no non-linearity association was observed (P_{non-linearity} = 0.11). The summary OR of bladder cancer for an increase of 1 cup/day of tea was 0.97 (95% CI 0.88–1.06, P_linear = 0.47). For case-control studies, we found no evidence of a non-linearity association between tea consumption and bladder cancer risk (P_{non-linearity} = 0.60). We then used the linear model among case-control studies. The pooled OR of bladder cancer for an increase of 1 cup/day was 1.02 (95% CI 0.98–1.06, P_linear = 0.27).

Black tea consumption was examined in 3 cohort studies (Heilbrun et al., 1986; Chyou et al., 1993; Nagano et al., 2000), 5 hospital-based case-control studies (Kunze et al., 1992; Wakai et al., 2004; Demirel et al., 2008; Hemelt et al., 2010; Wang J. et al., 2013), and 2 population-based case-control studies (Ohno et al., 1985; Nomura et al., 1991). No significant association was observed between black tea consumption and risk of bladder cancer (OR = 0.84, 95% CI 0.70–1.01) (Figure 4), with moderate between-study heterogeneity (I² = 34.9%, P_{heterogeneity} = 0.13). For cohort studies, the combined OR was 1.06 (95% CI 0.75–1.50), with low between-study heterogeneity (I² = 9.3%, P_{heterogeneity} = 0.33) (Table 2). For hospital-based case-control studies, the pooled OR was 0.79 (95% CI 0.58–1.08), with low to moderate between-study heterogeneity (I² = 42.8%, P_{heterogeneity} = 0.14). For population-based case-control studies, the combined OR was 0.80 (95% CI 0.62–1.02), with no between-study heterogeneity (I² = 0.0%, P_{heterogeneity} = 0.51).

Four case-control studies (Kunze et al., 1992; Wakai et al., 2004; Hemelt et al., 2010; Wang J. et al., 2013) were included for the dose-response meta-analysis of black tea consumption. The results had been shown in Figure 5. There was no evidence of a non-linearity association between black tea consumption and bladder cancer risk (P_{non-linearity} = 0.06). A linear model was used among these studies. The poled OR of bladder cancer for an increase of 1 cup/day of black tea was 0.89 (95% CI 0.76–1.05, P_linear = 0.18).

Green tea consumption was assessed in 3 cohort studies (Chyou et al., 1993; Nagano et al., 2000; Kurahashi et al., 2009), 3 hospital-based case-control studies (Wakai et al., 2004; Hemelt et al., 2010; Wang J. et al., 2013), and one population-based case-control study (Wilkens et al., 1996). No significant association was observed between green tea consumption and bladder cancer risk (OR = 0.95, 95% CI 0.73–1.24) (Figure 6), with moderate heterogeneity (I² = 52.5%, P_{heterogeneity} = 0.05). For cohort studies, the combined OR was 1.20 (95% CI 0.90–1.59), with no heterogeneity (I² = 0.0%, P_{heterogeneity} = 0.86) (Table 2). For hospital-based case-control studies, the pooled OR was 0.73 (95% CI 0.53–1.00), with moderate between-study heterogeneity (I² = 34.5%, P_{heterogeneity} = 0.22). For population-based case-control studies, the combined OR was 1.04 (95% CI 0.59–1.84).

Two cohort studies (Nagano et al., 2000; Kurahashi et al., 2009) and three case-control studies (Wakai et al., 2004; Hemelt et al., 2010; Wang J. et al., 2013) were included for the dose-response meta-analysis of green tea consumption. The results had been shown in Figure 7. No evidence of non-linearity association was detected between green tea consumption and bladder cancer risk (P_{non-linearity} = 0.92). A linear model suggested that the summary OR of bladder cancer for an increase of 1 cup/day of green tea was 1.02 (95% CI 0.94–1.1, P_linear = 0.66).

Mate consumption was investigated in 2 case-control studies (Bates et al., 2007; De Stefani et al., 2007). No significant association was observed between mate consumption and bladder cancer risk (OR = 2.17, 95% CI 0.70–6.74), with moderate to high between-study heterogeneity (I² = 75.0%, P_{heterogeneity} = 0.05) (Table 2).

Oolong tea consumption was examined in only one study (Lu et al., 1999). An elevated risk was observed between oolong tea consumption and bladder cancer risk (OR = 3.00, 95% CI 1.20–7.47) (Table 2).

Results of subgroup analyses had been shown in Table 2. For tea consumption and risk of bladder cancer, subgroup analyses defined by study design, sex, study location, adjustment for age, adjustment for smoking, and smoking status did not show any substantial change in the summary OR, with no evidence of between-group heterogeneity by meta-regression analysis (Table 2). For black tea consumption and risk of bladder cancer, subgroup analyses defined by study design, study location, adjustment for age, and adjustment for smoking did not show any substantial change in the summary OR, with no evidence of between-group heterogeneity by meta-regression analysis (Table 2). For green tea consumption and risk of bladder cancer, subgroup analyses defined by study design and study location did not show any substantial change in the summary OR, with no evidence of between-group heterogeneity by meta-regression analysis (Table 2). Sensitivity analysis by removing four studies (Morgan and Jain, 1974; Howe et al., 1980; Zeegers et al., 2001a; Demirel et al., 2008) with crude ORs did not change the overall results of tea consumption among all studies(OR = 0.99, 95% CI 0.89–1.10), cohort studies (OR = 1.02, 95% CI 0.84–1.23), hospital-based case-control studies (OR = 1.01, 95% CI 0.78–1.32), and population-based case-control studies (OR = 1.00, 95% CI 0.90–1.11),; but the heterogeneity was reduced in cohort studies (I² = 14.8%, P = 0.32 for heterogeneity) (Figure 8).

No evidence of publication bias was detected in the tea (P = 0.43 for Egger's test; Figure 9), black tea (P = 0.76 for Egger's test; Figure 10), and green tea (P = 0.06 for Egger's test; Figure 11) consumption.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The reviewer Q.D. and handling Editor declared their shared affiliation, and the handling Editor states that the process nevertheless met the standards of a fair and objective review.