Dietary B group vitamin intake and the bladder cancer risk: a pooled analysis of prospective cohort studies

Data were derived from the BLadder cancer Epidemiology and Nutritional Determinants study (BLEND): a large international consortium on dietary factors and BC risk, compromising a total of 11,000 cases and over 680,000 non-cases aged between 18 and 100 years from different countries in Europe, America, Asia and Australia [25]. Currently, BLEND consists of 19 case–control and seven cohort studies.

The present study pooled data from the BLEND cohort studies only. Four cohort studies, including a total of 2915 cases and 530,012 non-cases, had sufficient information on dietary B group vitamin intake to be eligible for inclusion in our analyses. Informed consent was obtained from all individual participants included in each study. Included studies were the Netherlands Cohort Study on diet and cancer (NLCS) [26], the Radiation Effects Research Foundation (RERF) atomic bomb survivors study [27], the VITamins And Lifestyle study (VITAL) [28] and the European Prospective Investigation into Cancer and nutrition (EPIC) [29, 30] (Table 1).

Details on the methodology of the BLEND consortium have been described elsewhere [25]. All included studies used a self-administered or trained interviewer administered food frequency questionnaire (FFQ) that was validated on either food groups [28, 30‐33], and/or energy intake [33, 34]. The period of recalling the dietary intake and the method used to validate the intake differed per study. In brief, (a) in the NLCS participants were asked to report on their dietary intake during the preceding year before study enrolment. This method was validated by a 5-year reproducibility test [35]; (b) the Vital study used a time reference for all dietary questions of “in the last 3 months” and used 24-h dietary recalls and a 4-day food record to validate the dietary intake of the participants [28]; (c) studies included in the EPIC study used either a 7-day food consumption diary or 24-h dietary recalls to validate the reported dietary intake of the participants during the preceding year [29]; (d) the RERF also used 24-h recalls to assess the bias and precision of their FFQ in which a 1 year time reference was used [32]. The collected dietary data were harmonized and categorized using the hierarchical Eurocode 2 food coding system (ECC) developed by the European Union, besides, weekly, monthly or yearly intake were converted to weekly food intake.

As a second step, all recoded food items were converted into nutrients using the United States Department of Agriculture (USDA) food composition database [36]. This database has been validated for B group vitamins [37]. For this, we chose the nutrient content per 100 g of generic food items where possible. Raw products were preferred over cooked/boiled for fruits, whereas for meat, fish, vegetables (except for salad vegetables—see below) and pulses roasted/cooked/boiled/grilled was preferred over raw. For example, for “Whitefish” we chose "Fish, whitefish, mixed species, cooked, dry heat" (code 15223) and for “Chicken breast” we chose “Chicken, broilers or fryers, breast, skinless, boneless, meat only, cooked, grilled” (code 05747). The nutrient content for endive, lettuce, fennel, chicory, garlic, chives, radish, cucumber, avocado, morel, and cress was selected as raw, for tomatoes the raw and cooked values were averaged, and for olives “canned” was selected.

Two approaches were used for food items not available in the USDA food composition database: (1) American terms were matched (i.e. the ECC is written in British English and the USDA database in American English). For example, “Courgette” was matched to zucchini; (2) if food items remained not found, we searched the internet for a description of the food item. For example, “Runner beans” were matched to kidney beans based on the information that kidney beans are the mature seed of the runner bean plant. For ECC pooled food items (e.g. salmon and trout), individual items were searched in the USDA database and the nutrient content was averaged.

The final nutrient intake was converted from weekly intake to daily intake (i.e. for each nutrient a nutrient in micrograms/day was created). If possible, portion sizes were adapted from individual studies, and otherwise based on USDA database information.

Person-years of follow-up for each participant was calculated from date of study enrolment until date of BC diagnosis, or date of ending follow-up (e.g. date of death, loss to follow-up, or study exit), whichever came first. For the NLCS study, a nested case–cohort approach was applied, in which the number of person-years at risk was estimated based on a sub-cohort that was randomly sampled [26].

Each study ascertained incident BC with International Classification of Diseases for Oncology (ICD-O-3 code C67) using population-based cancer registries, health insurance records, or medical records. The term BC was used for all urinary bladder neoplasms. BCs were classified into non-muscle invasive bladder cancer (NMIBC) and muscle invasive bladder cancer (MIBC). NMIBC included non-invasive papillary carcinomas confined to the urothelium (stage Ta), and carcinomas that invaded the lamina propria of the bladder wall (stage T1). High-grade flat non-invasive carcinomas confined to the urothelium (carcinoma in situ; CIS) without other concomitant tumour stages [i.e. T1/Ta (classified to non-muscle invasive prior) or MIBC] were also classified as NMIBC. MIBC included carcinomas that invaded into the detrusor muscle (stage T2), carcinomas that invaded into the peri-vesical tissue (stage T3), and carcinomas that invaded adjacent tissues and organs (most often the prostate or uterus, stage T4).

In addition to information on dietary intake, the BLEND data also included study characteristics (design, method of dietary assessment, recall time of dietary intake and geographical region), participant demographics (age, sex and ethnicity), BC pathology (MNIBC and MIBC), and smoking status (current/former/never and pack years), which were all measured at baseline [25].

The differences between the different exposure categories were examined by t test for continuous variables. To assess the association between dietary B group vitamins and BC risk, cox proportional hazard regression analysis was used to obtain hazard ratios (HRs) and corresponding 95% confidence intervals, stratified by study centre. Scaled Schoenfeld residuals were estimated for each covariate to test the proportional hazards assumption. In addition, the appropriateness of the use of the log-normal distribution was tested using a Wald test, and again no evidence of violation was found [38].

Separate analyses were undertaken for each B group vitamin compound individually, for all B group vitamins combined, and for combinations of B group vitamins based on their biological function. Combining of B group vitamin intakes was done by multiplying the individual compound intakes. B group vitamin intake was classified in tertiles (low/moderate/high intake) based on the distribution of all included study participants per study separately, since not every study measured an equal amount of food items of focussed on the same food groups. The Cox-regression model used low consumers as the reference group and was adjusted for previously shown BC risk factors: model 1: adjusted for the confounders age, sex (male/female), smoking status (was defined as a dummy variable: 0 (never smokers); 1 [current light smokers (i.e. smoking less than 20 pack-years)]; 2 [current heavy smokers (i.e. smoking more than 20 pack-years)]; 3 [current smokers (no information on pack-years)]; 4 [former light smokers (i.e. smokers who ceased smoking over 1 year prior and smoked less than 20 pack-years)]; 5 [former heavy smokers (i.e. smokers who ceased smoking over 1 year prior and smoked more than 20 pack-years)]; 6 [former smokers (smokers who ceased smoking over 1 year prior and no information on pack-years)])) [5, 6] and model 2: additionally, adjusted for water (low/high) (because of the water-soluble origin of B group vitamins [23]).

The Wald-test was used to derive the p value for linear trend. To understand the relevance of the effect modification, the main interaction terms between B group vitamin consumption (low/moderate/high) and sex, smoking and water were added to the model. P interaction < 0.05 was considered statistically significant where upon all analyses were stratified for the covariate of interest. In addition, sensitivity analyses were performed in which BC cases diagnosed within the first 5 years after recruitment were excluded.

Based on a priori hypothesis, additional supplemental analyses were performed by BC subtype (i.e. NMIBC and MIBC). In addition, dose–response analyses of vitamin intake (plotted on the x-axis) and HR (model 2, plotted on the y-axis) were performed using a nonparametric test for trend.

All statistical analyses were performed using Stata14 (StataCorp LLC, College Station, TX).

Baseline characteristics of the study population are described in Table 1. In total, 3265 cases and 580,634 non-cases were included in our analyses. Cases were on average older than non-cases (60.3 vs 52.6 years), more likely to be current or former smoker (37.58% and 42.02% vs 20.82% and 29.15%, respectively) and there were approximately three times more male than female cases (2189 vs 726). Compared to non-cases, cases consumed on average more vitamin B1 (1.65 mg vs 1.01 mg), B2 (3.63 mg vs 0.97 mg), B3 (16.83 mg vs 12.11 mg), B6 (2.05 mg vs 1.51 mg) and B9 (282.81 µg vs 185.59 µg).

Overall, no significant associations of higher intakes of any of the B group complex vitamins and BC risk were observed (Table 2), except for a moderate intake of vitamin B1, which showed to slightly increase the risk of BC (HR_B1: 1.10, 95% CI: 1.00–1.20) (Table 2). No other significant associations were observed.

After removing BC cases diagnosed within the first 5 years after recruitment, similar increased BC risks were observed (Supplementary Table 1).

Overall dose–response curves are shown in Fig. 1. A slightly increased BC risk was observed for higher intake of vitamin B1 (p = 0.03) and a decreased risk for higher intake of vitamin B12 (p = 0.002). No other compound or combination showed a significant association with BC risk.

For men, a significant increased BC risk was observed for high intake of the vitamins B1 (p = 0.009) and B3 (p = 0.02) (Fig. 2).

Among women, significant decreased risks were observed for high intake of the vitamins B3 (p = 0.02), B6 (p = 0.005), B9 (p = 0.03), B12 (p = 0.0001) and for the vitamins related to the reduction of oxidative stress (p = 0.02) (Fig. 3).

Experimental studies show that vitamin B1 is needed for the metabolism of glucose [41], thereby delivering the fuel of our cells [42]. Contrary, the present study showed a slight increased BC risk for moderate and high vitamin B1 intake among male, while among female high intake was associated with a decreased BC risk. No previous observational studies were conducted on the influence of vitamin B1 on BC risk, nonetheless, several attempts have been made to correlate vitamin B1 intake to other cancer types. However, results remain inconclusive [43, 44]. A possible explanation for the discrepant findings between male and female might be the different source from which participants retrieved their vitamin B1. While male mainly retrieved their vitamin B1 intake from meat consumption, the main vitamin B1 source for female are vegetables (Supplementary Table 2). Meat consumption has previously been associated with an increased BC risk [45‐50], while high vegetable consumption showed a decreased BC risk [9, 10]. Moreover, after adjustment of the main food resources (i.e. meat and vegetables) results for men and women were similar and did not reach statistical significance, thereby strengthening the hypothesis that indeed the different food source might explain the observed gender difference. However, future research is needed to further clarify this gender difference.

Vitamin B2 is a coenzyme for many metabolic processes in the body [18, 41]. Although this suggests an inverse effect of vitamin B2 on BC risk, in current observational studies this effect is controversial [51]. The present study also shows conflicting results; among men moderate vitamin B2 intake showed an increased BC risk, while among women a decreased risk was observed. This gender difference might be explained by the source. It is expected that men mainly derive vitamin B2 from meat, while women might derive it mainly from other sources. This is confirmed by the higher vitamin B2 meat-derived intake in men compared to women (M_{meat-B2 men}: 0.70 mg, SD = 0.63, M_{meat-B2 women}: 0.55 mg, SD = 0.50, t = 90.96, p < 0.001) (Supplementary Table 2). Meat has previously been associated with an increased BC risk due to its pro-carcinogenic components [45‐50], which could abolish the positive effect of vitamin B2. However, adjustment for the main food sources did not significantly change the findings. Therefore, future research is needed to clarify this gender difference.

Vitamin B3 plays a role in energy metabolism, which is essential to maintain cellular metabolism and respiration [52] and it is important for genetic and epigenetic regulators [52]. Studies of the consequences of DNA damage in cultured mouse and human cells as a function of vitamin B3 status have supported the hypothesis that vitamin B3 may be a protective factor in limiting carcinogenic events [53]. In line with this, we observed a decreased BC risk for high dietary vitamin B3 intake among women. Since meat is a rich source of vitamin B3 [54], this again could explain the observed gender differences. In our study, men retrieved on average more vitamin B3 from meat than women (M_{meat-B3, men}: 17.83 mg, SD = 15.71, M_{meat-B3 women}: 14.54 mg, SD = 13.44, t = 78.49, p < 0.001) (Supplementary Table 2). However, adjustment for the main food sources did not significantly change the findings. Therefore, future research is needed to clarify this gender difference. When analyzing the effect of the vitamins B1, B2 and B3 (i.e. the vitamins related to energy metabolism) together, moderate intake among men showed an increased risk on BC and high intake a decreased risk among women.

Oxidative stress is an imbalance in the cell which leads to DNA damage [55], thereby increasing cancer risk [39]. Experimental studies showed that the vitamins B6 and B2 are involved in oxidative stress reduction by catalyzing regulatory enzymes [55, 56]. Results of the present study confirm these findings by showing an inverse association of high intake of vitamin B6 on BC risk among women, in line with a previously conducted cohort study [23].

Vitamin B9 plays a pivotal role in cell metabolism which is necessary for DNA synthesis and repair, as well as for methylation [57]. Previous epidemiological studies and a meta-analysis confirm this protective role of vitamin B9 in the development of BC [14]. The present study also shows a reduced BC risk among women (for both vitamin B9 separate as well as combined with vitamin B12). In men, this protective effect could not be observed. This is possibly the result of the fact that men are less responsive to vitamin B9 than women, due to a lower body mass in which the folate dose distributes over a larger volume [58]. The difference might also be explained by androgens, which are involved in one-carbon metabolism [59]. In our study, men did consume on average more vitamin B9 than women (M_{B9, men}: 194.27 µg, SD = 128.66, M_{B9 women}: 182.12 µg, SD = 96.28, t = 37.76, p < 0.001).

Vitamin B12 is essential for the reduction of DNA damage [60]. Experimental studies show that vitamin B12 deficiency mimics radiation damage to DNA, possibly leading to cancer [61]. However, observational studies showed no effect of high intake of vitamin B12 on BC risk [62]. The present study only observed a significant protective effect of high vitamin B12 intake among women. The main sources of vitamin B12 is meat [54]. Since meat is mainly consumed by men (M_{B12, meat men}: 3.86 µg, SD = 4.30, M_{B12 meat women}: 3.35 µg, SD = 3.85, t = 43.03, p < 0.001) (Supplementary Table 2), it might be argued that the observed effect in men are caused by the earlier mentioned negative effects of other (micro)nutrients in meat [63]. However, adjustment for the main food sources did not significantly change the findings. Therefore, future research is needed to clarify this gender difference.

Although BLEND is one of the largest known pooled cohort studies investigating the association between dietary B group vitamin intake and the risk of developing BC, allowing for detailed analyses with enough statistical power, it has several limitations. First, limited information was available for possible BC risk factors, such as body mass index, physical activity, socioeconomic status, and occupational exposures. Nevertheless, current literature shows only a small proportion of BC cases can be attributed to these factors [23, 64‐66]. In addition, no information was available on comorbidities that may make people alter their diet [67], or of which the drugs may influence the bioavailability of B group vitamins in the body [68]. At last, information on alcohol was lacking which might influence B vitamins’ absorption [56].

A second limitation arises from the use of FFQs, which could lead to recall bias, systematic and random error when estimating vitamin intake. However, since the dietary intake of all included studies were validated, recall bias has likely only played a minor role. In addition, measurement error could be negligible, considering our large sample size.

Thirdly, although people are less likely to change their dietary habits at an older age, most of the included studies only measured their participants at baseline and we were, therefore, unable to take possible changes of dietary habits over time into account. This could have led to misclassification of long-term exposure [69]. However, the included NLCS study repeated the questionnaire 5 year after baseline, and showed only a minor decline in average intake for all food items [35].

Fourthly, most of the included studies did not provide information on supplement use. Therefore, we were unable to take supplemental vitamin intake into account, which may have led to an underestimation of the true effect of B group vitamins.

Fifthly, a single database was used for the conversion of food into nutrient intake. Since the food composition of similar food items and the food fortification may differ between different countries, the use of country specific food composition tables might be more accurate. Previous studies, however, showed that the use of a common food composition database advantages over the use of country-specific food composition databases in that errors are consistent between the countries [70]. In addition, all our main regression analyses were study centre stratified.

Besides, results obtained from cohort studies on diet and cancer risk cannot always rule out the possibility of reversed causality. Since there is no evidence that people are likely to alter their diet in the period before BC diagnosis, we decided to not exclude study participants who received a BC diagnosis within a short period of follow-up.

Finally, in view of multiple testing, it could indeed be debated whether, for instance, Bonferroni p value adjustments should have been applied. However, it previously has been argued that the use of Bonferroni p value adjustments is impractical and likely too conservative when testing a priori hypotheses [71]. Since we were able to formulate plausible a priori hypotheses regarding all the included analyses, based on data from previous studies, we did not apply Bonferroni correction in our analyses. Moreover, if we had adjusted for the number of main analyses being performed (n = 10) the significance level would have been 0.005. In that case, most of the observed associations between the vitamin B intake and BC risk would still be statistically significant.