Consumption of Vegetables and Fruits and Urothelial Cancer Incidence: a Prospective Study¹

Urothelial cancer is the seventh most common cancer among men, accounting for ∼200,000 new cases/year worldwide (1). Over the last 4 decades, many epidemiological studies and several reviews have been conducted to investigate determinants of urothelial cancer (2, 3, 4, 5, 6). These studies suggested that urothelial cancer is influenced by environmental factors, including cigarette smoking, fluid consumption, and exposure to industrial chemicals (e.g., aromatic amines) and diet.

Among the dietary factors, most epidemiological studies support a possible protective effect of vegetable (7, 8, 9, 10, 11, 12) and fruit consumption (7, 9, 12, 13, 14), although some studies have reported no association for vegetable (14, 15) or fruit consumption (8, 15). Previous studies also have been inconsistent with regard to which vegetables and fruits may be responsible for an inverse association.

Some possible anticarcinogenic mechanisms of substances in vegetables and fruits include antioxidant effects, effects on cell differentiation, increased activity of enzymes that detoxify carcinogens, blocked formation of nitrosamines, preserved integrity of intracellular matrixes, effects on DNA methylation, maintenance of normal DNA repair, increased apoptosis of cancer cells, and decreased cell proliferation (16). From a biological viewpoint, however, it is not yet clear whether the total amount of consumed vegetables and fruits is relevant for the possible protection against cancer or only the consumption of specific vegetables.

NLCS³ with 6.3 years of follow-up and with substantially more incident urothelial cancer cases available than in previous studies provided the opportunity to study prospectively detailed relationships between the total amount of vegetable and fruit consumption, as well as the consumption frequency of specific vegetables and fruits and urothelial cancer incidence for men and women consuming a Dutch diet.

This population-based prospective cohort study on diet and cancer started in the Netherlands in September 1986. In this year, 340,439 persons received a baseline questionnaire. The overall participation rate in the cohort was 35%, which led to 120,852 participants (58,279 men and 62,573 women aged 55–69 years). The study population originated from 204 municipal population registries throughout the country. The case cohort approach was used for data processing and analysis (17). Cases were enumerated from the entire cohort, whereas the accumulated person years in the cohort were estimated from a subcohort sample. After this approach, a subcohort of 3,500 subjects (1,688 men and 1,812 women) was randomly sampled from the cohort after baseline exposure measurement. The subcohort has been followed up for vital status information. No subcohort members were lost to follow-up during the follow-up period. The study design, including data collection strategies, has been described in detail previously (18).

Follow-up for incident cancer was established by record linkage to cancer registries and the Dutch national database of pathology reports (19). The completeness of cancer follow-up was estimated to be >95% (20). The presented analysis is restricted to cancer incidence in 6.3 years of follow-up, from September 1986 to December 1992. After excluding prevalent cases with cancer other than skin cancer, a total of 3,346 subcohort members (1,630 men and 1,716 women) and 619 incident cases (532 men and 87 women) with microscopically confirmed, incident carcinomas of the urinary urothelial, ureters, renal pelvis, or urethra were identified. Of these cases, 584 (94.3%) were diagnosed with urothelial cancer, of which 559 (95.7%) were transitional cell carcinomas.

At baseline, the cohort members completed a mailed, self-administered questionnaire on risk factors for cancer. Usual consumption of food and beverages during the year preceding the start of the study was assessed with a 150-item semiquantitative food frequency questionnaire. Participants were asked to report their consumption frequency of 21 vegetables (i.e., string/French beans, broad beans, Brussels sprouts, cauliflower, cabbage (white/green), kale, spinach, cooked endive, raw endive, lettuce, leek, onions, cooked carrots, raw carrots, sweet peppers, tomatoes, red beets, sauerkraut, mushrooms, gherkins, and rhubarb) and nine fruits (i.e., oranges, mandarins, grapefruits, orange/grapefruits juice, apples/pears, bananas, strawberries, grapes, and raisins/other dried fruits). Categories ranged from never or less then once per month to “three to seven times per week” for vegetable consumption and from never or less than once per month to “six to seven times per week” for fruit consumption. For onions, tomatoes, sweet peppers, and mushrooms, participants were asked to report the number they usually ate per week (onions and tomatoes) or month (sweet peppers and mushrooms, in boxes of 250 grams). Frequency of vegetable consumption was asked separately for summer and winter, from which a combined consumption frequency was calculated. For string beans and cooked endive, portion sizes were asked, which were used as indicators to calculate portion sizes of other vegetables, according to a vegetable-specific algorithm. To derive an individual serving size for each type of vegetable, the indicator serving size was multiplied with a type-specific factor calculated from the same pilot study data as the ratio of the means of the specific to the indicator serving sizes. The frequencies and amounts with regard to fruit consumption have been converted to consumption in grams per day using standard portion sizes. The choice of items for inclusion in the questionnaire was such that it could predict most of the interindividual variation in nutrient intakes. These items covered almost all vegetables and fruits eaten regularly in 1986, with the exception of chicory, red cabbage, cucumber, and broccoli, although the consumption of broccoli was not frequent at baseline. An open-ended question on other foods eaten on a regular basis was also included. Participants could write down how many times per week they ate such a food and how much they ate on each occasion. The questionnaire has been validated against a 9-day diet record. The Spearman correlation coefficient was 0.38 and 0.60 for total vegetable and total fruit consumption, respectively (21).

Incidence RRs and corresponding 95% CIs for urothelial cancer were estimated using exponentially distributed failure time regression models (22) with the Stata statistical software package (23). SEs were estimated using the robust Hubert-White sandwich estimator to account for additional variance introduced by sampling from the cohort. This method is equivalent to the variance-covariance estimator presented by Barlow (24) and Lin and Ying (25). We confirmed constancy of the baseline hazard visually by plotting the natural logarithm of the baseline survival function against failure time.

The following variables were subsequently considered as potential confounders based on earlier analyses (26, 27, 28, 29): age (years); consumption of alcohol (grams/day); coffee (ml/day); tea (ml/day); total water (ml/day); current cigarette smoking (yes/no); cigarette smoking amount (cigarettes/day); cigarette smoking duration (years of cigarette smoking); occupational exposure to dye, rubber, leather, or vehicle fumes (ever/never); and first-degree family history of urothelial cancer (yes/no). Those variables that (combined) showed a >10% influence on the risk of urothelial cancer when considered in a multivariable model were included as covariates in multivariable analyses.

Subjects with incomplete or inconsistent dietary data were excluded from analyses (21). To check the quality and consistency of the responses on the several vegetable questions, a specific error index was computed. When the error index exceeded a certain value, i.e., more than three errors, subjects were excluded from analyses on vegetable consumption. The error index was calculated as the sum of variables that indicated each the presence of an inconsistency or other response error. Finally, a total of 538 urothelial cancer cases (463 men and 75 women) and 2953 subcohort members (1456 men and 1497 women) remained for analyses on vegetable data, and 569 urothelial cancer cases (491 men and 78 women) and 3123 subcohort members (1525 men and 1598 women) remained for analyses on fruit data.

The RRs concerning vegetable and fruit consumption were similar for men and women. Therefore, we conducted regression analyses for men and women combined (Ps for interaction >0.40) We first evaluated combined vegetable and fruit consumption, total vegetables, prepared and raw vegetables, vegetables categorized in botanical groups (i.e., leafy vegetables, pulses, brassicas, and allium vegetables), total fruits, and citrus fruits after categorizing subjects into quintile or tertile levels of consumption depending on the distribution in the subcohort. Cases detected during the first and the first 2 years of follow-up were excluded from analyses to evaluate whether preclinical symptoms might have influenced the results. Specific vegetables and fruits were tested for linearity and evaluated as continuous variables (grams/day), and their multivariable RRs were expressed per increment of 25 grams/day, corresponding to a consumption frequency of a vegetable approximately once per week. The independent contribution of each specific vegetable or fruit was assessed by an analysis in which total vegetable and fruit consumption was included in the model simultaneously. Tests for trend in the RRs were based on likelihood ratio tests. RRs for quintiles of vegetable and fruits consumption were also computed separately in strata of smoking status and amount. For the same variables, subgroup analyses were performed after stratification by tumor morphology and invasiveness. To ensure that the results were not influenced by changes in exposure by subjects with preclinical urothelial cancer, we conducted analyses with and without cases diagnosed in the first 1 or 2 years of follow-up. The results from these analyses were essentially similar and, therefore, analyses with all cases were presented.

Mean daily consumption of vegetables and fruits was calculated for categories of the most important potential confounders (Table 1). Age, coffee consumption, current cigarette smoking, cigarette smoking amount, cigarette smoking duration, high-risk occupation, and family history of urothelial cancer were not associated with vegetable consumption, whereas alcohol, tea, and total water consumption were positively associated. Concerning fruit consumption, we found no association with age, high-risk occupation, and family history of urothelial cancer; inverse associations were found with alcohol consumption, coffee consumption, current cigarette smoking, cigarette smoking amount, and cigarette smoking duration, and positive associations were found with tea consumption and total water consumption. Vegetable and fruit consumption were positively associated with each other (Table 1).

Incidence RR estimates for quintiles, quartiles, or tertiles of consumption are presented in Table 2 for total vegetable and fruit consumption and the consumption of groups of vegetables and fruits. Despite the distributions of vegetable and fruit intake across potential confounders, only age, sex, cigarette smoking amount, and cigarette smoking duration appeared to be the most influential confounders in multivariable analyses. The age- and sex-adjusted incidence RR showed an inverse association between the combined consumption of vegetables and fruits and urothelial cancer risk, which disappeared after additional correction for cigarette smoking amount and cigarette smoking duration. Of all vegetable groups, there was only suggestive evidence for an inverse association between the consumption of brassicas and urothelial cancer risk (RR: 0.75; CI: 0.54–1.04, comparing extreme consumption quintiles), although the linear dose response trend was borderline statistically significant (P trend = 0.06). Total fruit consumption was inversely associated (RR: 0.74; CI: 0.53–1.04; P trend = 0.02), whereas the consumption of citrus fruits did not appear to be clearly associated with urothelial cancer risk (RR: 0.85; CI: 0.62–1.17; P trend = 0.22), comparing extreme consumption quintiles. Additional adjustment for cigarette smoking amount, cigarette smoking duration, total fruit consumption (for vegetable items), total vegetable consumption (for fruit items; Table 2), or other potential confounders did not change the results essentially.

For specific vegetables and fruits, RRs are presented for increments in mean daily consumption of 25 grams (Table 3). Statistically significant inverse associations were found for the consumption of cauliflower (RR: 0.77; CI: 0.61–0.98), cooked carrots (RR: 0.66; CI: 0.47–0.92), and mandarins (RR: 0.63; CI: 0.42–0.96), although the range in consumption was small (2–28, 0–20, and 0–15 grams/day, respectively). No statistically significant positive associations were observed. Table 3 also gives estimates for each vegetable or fruit after simultaneous adjustment for total vegetable and fruit consumption to assess their independent effects. The risk estimates remained practically stable (Table 3).

In Table 4, the age-, sex-, and smoking-adjusted RRs are presented for combinations of total vegetable and fruit consumption, both divided in tertiles of daily consumption. The protective effect of fruit consumption seemed largest among those in the lowest tertile of vegetable consumption. Subjects in the median tertiles of vegetable and fruit consumption had the lowest risk of urothelial cancer (RR: 0.66; CI: 0.43–0.95, compared with those in the lowest tertiles of vegetable and fruit consumption), although this could be attributable to chance (Table 4).

Cigarette smoking is an important risk factor for urothelial cancer. Therefore, the analyses were stratified on smoking status and smoking amount (Table 5). Total vegetable consumption was not associated with urothelial cancer risk across different cigarette smoking categories. The protective effect of fruit consumption was most pronounced in ex-smokers who smoked >15 cigarettes/day and in current smokers, regardless of smoking amount (Table 5).

It appeared that a protective effect of vegetable and fruit consumption was mainly confined to invasive transitional cell carcinomas. The RRs for noninvasive tumors did not indicate a clear inverse association for vegetable or fruit consumption (data not shown).

The results showed that total vegetable consumption was not associated with urothelial cancer risk, whereas the consumption of brassicas, especially cauliflower, was inversely associated with urothelial cancer risk. Although we did not find a clear inverse association between citrus fruits and urothelial cancer risk, the data suggested that an increased consumption of mandarins might protect against bladder cancer. The consumption of carrots might also decrease the risk of urothelial cancer.

The prospective nature of this cohort study together with completeness of follow-up, as has been achieved in this study, reduced the potential for selection bias to a minimum. Information bias is also largely avoided because dietary habits were reported before urothelial cancer was diagnosed. A change in dietary habits of subjects with latent urothelial cancer at the time of completing the baseline questionnaire remains, however, a possibility, although this is much less likely compared with subjects with, e.g., gastrointestinal cancer.

A potentially more realistic threat to the interpretation of the observed inverse associations is residual confounding by risk factors for urothelial cancer. The most important risk factor in this respect is cigarette smoking (26). We modeled cigarette smoking habits such that they best explained urothelial cancer risk, resulting in a model including the habitual number of cigarettes smoked per day (smoking amount) and the number of years smoked (smoking duration), both as continuous variables. We were not able to explain our results on the basis of confounding by other factors.

Could the measurement of vegetable or fruit consumption itself have biased the results? Vegetables are generally considered as food items that are not very easy to assess in food-frequency questionnaires (as well as in other methods of dietary assessment), particularly if portion sizes have to be estimated. In the NLCS validation study, the correlation coefficient for total vegetable consumption was 0.38 (21), which is rather low, but comparable with the figure reported for many other prospective studies (31). One of the reasons for this low correlation may be the relative lack of true contrast in the frequency of vegetable consumption in a population such as the Dutch, because people are accustomed to a diet including one hot meal per day, which usually includes vegetables. A consequence of a relatively large measurement error would tend to attenuate the estimated RR. Because of individual preferences, contrast in consumption frequency of many specific vegetables is much higher. It was not possible to assess validity for specific vegetables in the NLCS validation study, because 9 days of dietary record are not sufficient to estimate consumption frequency of specific vegetables. To minimize the amount of uninformative data in addition to the general dietary exclusion criteria, we excluded subjects who appeared not to have understood how to fill in the questions on vegetable consumption; an extreme score on the vegetable error index defined those subjects.

To our knowledge, five case-control studies (7, 9, 11, 15, 32) and six cohort studies (8, 10, 12, 13, 14, 33) have evaluated the relationship between total vegetable or fruit consumption and urothelial cancer risk. Most studies found inverse associations for total fruit consumption (7, 9, 12, 13, 14), total vegetable consumption (7, 8, 9, 10, 11, 12), and total consumption of fruits and vegetables combined (8, 14, 32), although one case-control study (15) and one cohort study (14) did not find an association between total vegetable consumption and urothelial cancer risk, and one case-control study (15) and one cohort study did not find an association between fruit consumption and urothelial cancer risk (8). Recently, a meta-analysis on epidemiological studies reported elevated urothelial cancer risks among subjects with low fruit consumption (RR = 1.40; CI: 1.08–1.83) and low vegetable consumption (RR = 1.16; CI: 1.01–1.34; Ref. 29).

One case-control study (34) and one cohort study (8) examined the association between brassica consumption and urothelial cancer risk. The results from both studies were in accordance with our results, indicating an inverse association. As in our study, the cohort study also presented an inverse association between cauliflower consumption and urothelial cancer risk (35). Brassicas have been widely studied for their anticarcinogenic properties in experimental studies and have been associated with reduced cancer risk (8, 16, 36, 37). They are unique in their high content of dithiolethiones and isothiocyanates. These organosulfur compounds have been shown to increase the activity of enzymes involved in the detoxification of carcinogens (8, 16).

Two case-control studies (34, 38) have reported a significant inverse association between carrot consumption and urothelial cancer risk, which is in agreement with the results of the present study, but three other cohort studies (8, 14, 39) have reported no association between orange-yellow vegetable consumption and urothelial cancer risk. These vegetables (e.g., carrots) are sources of α- and β-carotene, which are antioxidants (16).

To date, no study estimated the association between the consumption of citrus fruit (e.g., mandarins) and urothelial cancer risk. Citrus fruits are known for their high content of vitamin C, which, as an antioxidant, may protect cell membranes and DNA from oxidative damage. Vitamin C may additionally help prevent cancer via its ability to scavenge and reduce nitrite, thereby reducing substrate for the formation of nitrosamines.

The antioxidant mechanism has received most attention. The lack of effect of vegetable and fruit consumption among nonsmokers is consistent with this hypothesis because never smokers have not been exposed to smoking-induced oxidative stress.

Recently, Cohen (40) suggested that transitional cell carcinoma of the urinary urothelial might represent two distinct diseases: papillary transitional cell carcinoma, which is usually noninvasive, and the nonpapillary type, which tends to be invasive. In addition to different prognostic and pathogenic properties, these two diseases appear to involve different molecular events (40). From an etiological point of view, some distinction might be important. An inverse association between both vegetables and fruits was most pronounced in invasive transitional cell carcinomas of the urinary urothelial, irrespective of tumor morphology. Additional research is needed to evaluate this matter.

In conclusion, the data are suggestive of an inverse association between the consumption of brassicas, total fruit, and bladder cancer risk, which are unlikely to be attributable to residual confounding of smoking.

We thank the regional cancer registries, the Dutch national database of pathology reports, and the National Health Care Information Center for providing incidence data and S. Van de Crommert, H. Brants, W. van Dijk, M. Moll, J. Nelissen, C. de Zwart, A. Pisters, H. van Montfort, R. Schmeitz, T. van Montfort, and T. van Moergastel for assistance.