Renal cell carcinoma (RCC) accounts for 3 % of adult malignancies in the USA, and the incidence has been increasing in the USA for the last 30 years, with annual increments of 1·6 and 1·7 % in white men and white women(Reference McLaughlin, Lipworth, Tarone, Schottenfeld and Fraumeni1). In 1990, rates of RCC were 12 and 5 per 100 000 among white men and women(Reference Devesa, Silverman and McLaughlin2). Recent rates (2005) are reported as 18 and 9 per 100 000, respectively(Reference Moore, Wilson and Campleman3). The increase cannot be fully explained by early detection of pre-symptomatic tumours(Reference McLaughlin, Lipworth, Tarone, Schottenfeld and Fraumeni1). The reported ongoing epidemic of obesity in the USA(Reference Flegal4) and/or the increase in hypertension(Reference Fields, Burt and Cutler5) and diabetes(Reference Boyle, Thompson and Gregg6) may explain part of this increase, which occurred despite a drop in smoking rates(7, Reference Parker, Cerhan and Janney8). Although smoking(7, Reference Parker, Cerhan and Janney8), obesity(Reference Bergstrom, Hsieh and Lindblad9–Reference Pan, DesMeules and Morrison12), hypertension(Reference Brock, Gridley and Lynch10, Reference Chow, Gridley and Fraumeni11, Reference Grossman, Messerli and Boyko13) and diabetes(Reference Zucchetto, Dal Maso and Tavani14) have consistently been associated with RCC risk, few studies have tried to assess the association of decreased dietary intake of fruit and vegetable intake, taking into account constituent forms of fibre and other micronutrients, as well as assessing for interaction with sex, age and smoking(Reference Galeone, Pelucchi and Talamini15–Reference Hu, La Vecchia and Negri17). An increase in lipid peroxidation may partially explain some of the reason for increasing RCC risk(Reference Gago-Dominguez and Castelao18–Reference Greenland, Gago-Dominguez and Castelao20). To evaluate the association of dietary intake of fruits, vegetables and different types of fibre and other micronutrients with risk of RCC, we analysed RCC dietary data, along with other established and potential risk factors collected as part of a large population-based case–control study.
Material and methods
Study sample
A population-based case–control study of RCC and cancers of five other anatomic sites was conducted in Iowa between 1986 and 1989. Detailed methods are reported elsewhere(Reference Parker, Cerhan and Janney8, Reference Brock, Gridley and Chiu21, Reference Cantor, Lynch and Johnson22). Briefly, eligible cases were residents of the state of Iowa, aged 40–85 years, newly diagnosed with histologically confirmed RCC (ICD-O code 189.0) during July 1985–December 1987, and without a previous diagnosis of a malignant neoplasm. Cases were identified by the State Health Registry of Iowa(Reference Lynch, Logsden-Sackett and Edwards23). An introductory letter was followed by a telephone call in which potential participants were invited to complete a mailed questionnaire, designed either for direct respondents or their proxies, sent per request during the telephone contact. Of the 463 eligible RCC cases, questionnaires were completed for 406 (87·7 % response rate). Among these, 287 subjects completed the questionnaire designed for direct respondents and 119 completed a proxy questionnaire. An early version of the direct-respondent questionnaire, which did not include a question about possible proxy status, was completed by eighty-one of the 287 ‘direct questionnaire’ respondents. In the present analysis, these respondents were assumed to be the study subject since almost all of the 206 respondents who completed the later version of the direct respondent's questionnaire that asked about possible proxy status, were study subjects. Both versions asked the same questions on food consumption.
Controls were frequency-matched to all cases in the overall study by sex and 5-year age group. Controls, like cases, had to be without previous diagnosis of a malignant neoplasm. Controls under 65 years of age were selected randomly from computerised State of Iowa driver's license records, whereas controls aged 65 years and older (65+) were selected randomly from lists of Iowa residents provided by the USA Health Care Financing Administration (now the Centers for Medicare and Medicaid Services). Both sampling frames have been shown to achieve greater than 95 % coverage of the intended population(Reference Hartge, Cahill and West24). Of the 999 eligible controls under age 65 years, 817 (82 %) participated by returning a completed questionnaire; of 2036 eligible controls aged 65+ years, a total of 1617 participated (79 %). Among the 2432 control subjects sent direct-respondent questionnaires, 2064 were completed by the subject, 241 by a proxy and 127 by an undetermined respondent (assumed to be a direct respondent, as described).
The study was approved by the Institutional Review Boards of the USA National Cancer Institute and the University of Iowa.
Data collection
Data were collected by means of a self-administered mailed questionnaire, supplemented by a telephone interview where necessary. The questionnaire included information on demographics, anthropometric measures (weight history and adult height), usual non-occupational physical activity, smoking history, occupational history, past medical history (including self-report of physician-diagnosed hypertension and history of bladder/kidney infection), history of cancer among first-degree relatives and other factors. Of the 2434 controls, 607 did not have sufficient dietary data for analysis. A total of sixty-six controls were missing information on BMI and/or a history of hypertension. Of the 406 RCC cases, eighty-three did not have sufficient dietary information and ten did not have BMI and/or hypertension information. These subjects were excluded, leaving 323 cases and 1827 controls for the dietary analysis. Most of the 607 controls and eighty-three cases who were excluded due to insufficient dietary information had responded to a truncated telephone questionnaire that did not include diet.
Dietary analysis
Usual adult dietary intake was gathered with a FFQ that asked about the number of times per d, week, month or year (or rarely/never) of consumption for each of fifty-five food items, excluding dietary changes in the previous couple of years. Intake per d for each item was calculated and these data were summed to derive frequency of intake within each food group. Estimates of usual intake were derived for individual food items by multiplying the frequency of consumption of each item by an average serving size for males and females, separately, obtained from the National Health and Nutrition Examination Survey II (NHANES II)(Reference Dixon, Zimmerman and Kahle25, Reference Dresser26). Nutrients were then estimated by multiplying the intake of these foods by nutrient values derived from the United States Department of Agriculture food composition tables(Reference Dresser26) and a USDA-National Cancer Institute food composition database(Reference Dixon, Zimmerman and Kahle25). Adjustment for total food intake was carried out by the nutrient density method(Reference Hu, Stampfer and Rimm27). Each nutrient was individually divided by the subject's total energy intake before quartiles of intake were calculated. When nutrients were analysed, total energy consumption in kJ (continuous variable) was entered into a logistic regression model along with the other potential confounders. Two statistical packages were used: Statistical Package for the Social Sciences (version 11; SPSS, Inc.) and EPICURE (EPICURE, Inc.)(Reference Preston, Lubin and Pierce28).
Multiple logistic regression analysis was used to adjust for confounding by age (continuous), sex, smoking (eight categories of smoking duration and amount, respectively (based on distribution in controls), and smoking status), BMI at age 40 years, history of high blood pressure (yes, no), proxy status of respondents (direct or proxy respondent), alcohol intake(Reference Greving, Lee and Wolk29, Reference Lee, Hunter and Spiegelman30) and fatty spreads consumption(Reference Brock, Gridley and Chiu21). The maximum likelihood estimate of the OR, with 95 % CI, was used as the measure of association between either high food group intake or macro- or micronutrient intake and RCC(Reference Breslow and Day31). Tests for the trend across quartiles were performed by assigning the mean value of each respective quartile to the score variable and then testing the linear trend using a likelihood ratio test(Reference Breslow and Day31). Interactions between each variable (age, sex, smoking, hypertension and obesity) and the fruit- and vegetable-intake variables for RCC risk were tested by the likelihood ratio test(Reference Breslow and Day31) by comparing the log-likelihoods of logistic regression models with and without additional multiplicative terms for the interactions.
Results
Compared with controls, cases were somewhat younger and were more likely to be current smokers (OR 1·5; 95 % CI 1·1, 2·2), overweight or obese at age 40 years (OR 1·4, 95 % CI 1·1, 1·8), to report a history of hypertension (OR 1·8, 95 % CI 1·2, 2·4), to drink less alcohol (OR for more than two drinks/d 0·4, 95 % CI 0·3, 0·6), to consume more fatty spreads (OR 2·0, 95 % CI 1·3, 3·0) and to differ by respondent status (proxy; Table 1)(Reference Brock, Gridley and Lynch10, Reference Brock, Gridley and Chiu21). Therefore, these variables were included as confounders in subsequent analyses. Neither physical activity, coffee/tea consumption, education, family history of kidney cancer, nor history of kidney infection were risk factor and thus these factors were not included as covariates in any of the models. Among direct and proxy respondents, OR for smoking, obesity and hypertension, alcohol use and high fat consumption followed similar patterns (P interaction>0·5; data not shown)(Reference Zucchetto, Dal Maso and Tavani14).
Table 1 Demographic and life-style risk factors: Iowa case–control study of renal cell cancer
(Number of cases, number of controls, percentages, odds ratios and 95 % confidence intervals)
* Adjusted for age, sex, proxy status, years of smoking, number of cigarettes smoked per d, never/ever smoke, BMI age 40 years, blood pressure, alcohol consumption, fat consumption and energy where relevant.
We compared energy and percentage contribution of fat, protein and carbohydrate, by sex and case–control status, in our data with that in the NHANES II, which includes a nutritional survey conducted approximately contemporaneously(Reference Hartge, Cahill and West24). This was done as no validation studies were available from 1986 and we wanted an indication of the generalisability of our data to the general US population at the time. The dietary composition of total energy and distribution of macronutrients among both male and female controls from this study in Iowa was remarkably similar to the NHANES II study sample. In both populations, men consumed approximately 8000 kJ/d, of which fat comprised almost 40 % and women consumed approximately 5550 kJ/d, of which fat comprised about 35 %(Reference Brock, Gridley and Chiu21).
Table 2 presents associations between RCC risk and vegetables and fruits, either by food group, fibre nutrient or micronutrients in the total population; OR for vegetables and fruits either by food group, fibre nutrient or micronutrients in direct respondents followed similar patterns (data not shown as P interaction = 0·84). Intake of vegetables was the only food group associated with a decreased risk of RCC (OR 0·5; 95 % CI 0·3, 0·7; P trend = 0·002) (for the top quartile compared to the bottom quartile of intake).
Table 2 Associations between fruit and vegetables and renal cell cancer risk (food groups and nutrients) in the total population and stratified by age and smoking
(Odds ratios and 95 % confidence intervals)
* Adjusted for age, sex, proxy status, years of smoking, number of cigarettes smoked per d, never/ever smoke, BMI age 40 years, blood pressure, alcohol consumption, fat consumption and energy.
When intake of individual fibre constituents was investigated, only vegetable fibre intake was independently associated with decreased risks (OR 0·4; 95 % CI 0·2, 0·6; P trend < 0·001), but not fruit fibre OR 0·7; 95 % CI 0·4, 1·1) or grain fibre (OR 1·0; 95 % CI 0·6, 1·5).
β-Cryptoxanthin and lycopene were also associated with decreased risks, but when both were included in a mutually adjusted backwards model, only β-cryptoxanthin remained significant (OR 0·5; 95 % CI 0·3, 0·8; P trend = 0·01; data not shown in Table 2).
When fibre groups and nutrients were mutually adjusted for each other (in models that included other confounders), only consumption of vegetable fibre and β-cryptoxanthin remained significantly associated with lower RCC rates (OR 0·6, 95 % CI 0·3, 0·9, P trend = 0·03; OR 0·5, 95 % CI 0·3, 0·9, P trend = 0·02), respectively (for the top quartile compared to the bottom quartile of intake; data not shown in Table 2).
There was interaction between risk of RCC and vegetables and fruits either by food group or by micronutrients with two subgroups: smoking (P interaction cruciferous × smoking = 0·03) and age (P interaction β-cryptoxanthin × age = 0·007); there were no significant interactions with BMI, hypertension or sex (Table 2).
Thus in Table 2, the associations between RCC risk and these food groups and macro- and micronutrients are presented not only in the total population but also stratified by age and smoking. In those 65+ years of age, there was a significant negative association between RCC risk and intake of vegetable fibre, folate, vitamin C and β-cryptoxanthin. In non-smokers, we also found associations between RCC risk and higher intake of the fruit food group, cruciferous vegetables and fruit fibre (OR 0·3, 95 % CI 0·2, 0·6, P trend = 0·001; OR 0·5, 95 % CI 0·3, 1·0, P trend = 0·02; OR 0·4, 95 % CI 0·2, 0·8, P trend = 0·002), respectively (top compared to the bottom quartile of intake (Table 2)). When micronutrients were investigated, intake of both vitamin C and β-cryptoxanthin was associated with RCC among non-smokers but not smokers.
When nutrients were mutually adjusted in a stepwise regression model by subgroups, β-cryptoxanthin was the only one that remained associated with lower RCC risk among those aged 65+ years (OR 0·4; 95 % CI 0·2, 0·6; P trend < 0·001), and among non-smokers (OR 0·4; 95 % CI 0·2, 0·8, P trend = 0·002) (top compared to the bottom quartile of intake; data not shown in Table 2). Similar risks were seen when analyses were limited to direct respondents (P interaction>0·5).
Discussion
Results from this population-based, case–control study provide evidence for a link between high dietary intake of vegetables and a decreased risk of RCC. As decreased risks were also associated with increased vegetable intake, the individual fibre constituents and micronutrients were also investigated. Once the effect of dietary energy and fat consumption was taken into account, vegetable fibre, but not fruit and grain fibre, was significantly associated with decreased RCC risk. Vegetable fibre and β-cryptoxanthin showed the strongest association with RCC risk after mutual adjustment of all variables. These associations of low RCC risk with high intake of vegetable fibre and the micronutrient β-cryptoxanthin were also seen in those aged 65+ years and in non-smokers.
Our findings of a significant effect of vegetable intake are consistent with both past and recent large case–control, cohort and pooled studies. Our data showing an association for food groups are similar to those of Canadian(Reference Handa and Kreiger32), Italian(Reference Bravi, Bosetti and Scotti33) and US(Reference Grieb, Theis and Burr34) case–control studies. An Italian case–control study (with hospital controls) reported a significant two-fold association, similar to ours(Reference Talamini, Baron and Barra35). Out of thirteen case–control(Reference Handa and Kreiger32–Reference Yuan, Gago-Dominguez and Castelao44) and six cohort studies(Reference Bertoia, Albanes and Mayne45–Reference Weikert, Boeing and Pischon50), all case–control studies, three(Reference Rashidkhani, Lindblad and Wolk48, Reference van Dijk, Schouten and Kiemeney49, Reference Boffetta, Couto and Wichmann51) of the five large cohorts, and a large pooled analysis of thirteen cohort studies(Reference Lee, Mannisto and Spiegelman52) reported an association of vegetable intake with a decrease in RCC risk.
Our data also showed an association with cruciferous vegetables among non-smokers. In a pooled case–control study from four countries(Reference Wolk, Gridley and Niwa43) and in a Californian study(Reference Yuan, Gago-Dominguez and Castelao44), cruciferous vegetables were also found to be protective. Our finding of selected types of dietary fibre as the major nutrient associated with RCC risk is in accordance with the two studies which investigated the role of macronutrients, where fibre(Reference Hu, La Vecchia and DesMeules16, Reference Hu, La Vecchia and Negri17) was investigated.
When all food groups and types of fibre were entered in the same logistic model, vegetable fibre and β-cryptoxanthin remained as the micronutrients associated with inverse associations with RCC risk in our study. This result is consistent but more marked than that reported by Galeone et al. (Reference Galeone, Pelucchi and Talamini15) who investigated fibre constituents and found vegetable fibre to be significant (OR 0·73; 95 % CI 0·54, 0·97), but not fruit fibre (OR 1·01; 95 % CI 0·76, 1·34).
It is interesting that the only nutrient that was significantly associated with RCC risk in the pooled study of cohorts(Reference Lee, Mannisto and Spiegelman52) was α-carotene; however, other carotenoids were close to significance (β-carotene, β-cryptoxanthin and lycopene); the same carotenoids were also found to be associated with RCC risk in a large Canadian study and a US case–control study(Reference Hu, Mao and White39, Reference Yuan, Gago-Dominguez and Castelao44) but not in an Italian case–control study(Reference Bosetti, Scotti and Maso53). No association was observed for lycopene in these three studies(Reference Hu, Mao and White39, Reference Yuan, Gago-Dominguez and Castelao44, Reference Bosetti, Scotti and Maso53). Unlike the findings of Hu et al. (Reference Hu, La Vecchia and DesMeules16), the effect of vegetable fibre on RCC risk in our study remained reduced but significant after mutual adjustment with β-cryptoxanthin. We did not find any interaction with obesity or hypertension and nutrients with respect to RCC risk. Some other studies(Reference Yuan, Gago-Dominguez and Castelao44) have also found β-cryptoxanthin to be inversely associated with RCC risk, with effects stronger among non-smokers, as we observed. However, the association is not consistent among studies, and other investigations have not observed an association(Reference Hu, La Vecchia and Negri17, Reference Bosetti, Scotti and Maso53, Reference van Dijk, Schouten and Oosterwijk54). β-Cryptoxanthin and lycopene are found in a variety of fruit and vegetables such as oranges and tomatoes. In our population, these micronutrients were derived primarily from orange juice and tomato paste consumption and thus are significantly correlated with α-carotene, β-carotene and lutein (r 2 0·3 in all three, P < 0·05).
Results for micronutrients from individual cohort studies have been mainly null (with the exception of a finding in men in the USA)(Reference Lee, Hunter and Spiegelman30), with most showing no effects of individual carotenoids except when stratified by genotype. A study in the Netherlands found no effects of micronutrients(Reference van Dijk, Schouten and Oosterwijk54) except for an association of α-carotene and β-cryptoxanthin and folate with RCC risk in carriers of the wild-type gene for Von-Hippel Landau (VHL) tumours(Reference van Dijk, Schouten and Oosterwijk54).
In a large multicentre case–control study from Central and Eastern Europe, vegetable intake was found to be modified by three key folate metabolism genes(Reference Moore, Hung and Karami55). Blood levels of folate were found to be inversely associated with RCC risk in a cohort of Finnish male smokers(Reference Gibson, Weinstein and Mayne56), but interestingly, no dietary nutrient effects of any carotenoids or folate or fibre were observed in this cohort(Reference Bertoia, Albanes and Mayne45).
It is interesting that on subgroup analysis of non-smokers, fruit and vitamin C were also related to RCC risk, which has also been noted by others(Reference Hu, La Vecchia and Negri17, Reference Yuan, Gago-Dominguez and Castelao44). Whether this is due to consumption patterns of non-smokers (i.e. a healthier diet) or a real biological effect, needs to be elucidated; and this requires more work in a larger cohort of non-smokers.
A recently proposed putative mechanism that may shed light on these findings is the ‘lipid peroxidation hypothesis’. This mechanism not only explains the positive effects of smoking and fat on RCC risk, but also explains the associations of dietary antioxidants with kidney function. This hypothesis is supported by observations in both experimental chemically induced models and human renal cell tissue(Reference Gago-Dominguez and Castelao18, Reference Gago-Dominguez, Castelao and Yuan19).
Strengths of our present study include the use of a well-established tumour registry to ascertain cases(Reference Ries, Harkins and Krapcho57), a randomly selected control sample representative of the general population and high participation rates among cases and controls. In addition, we assessed external validity by comparing energy and percentage contribution of fat, protein and carbohydrate, by sex and case–control status, in our data with that in the NHANES II. The dietary composition of total energy and distribution of macronutrients among both male and female controls from this study in Iowa was remarkably similar to the NHANES II study sample. Additional strengths were our ability to investigate dietary fibre and to adjust for a wide variety of potential confounding factors including fat intake, which had a high prevalence among our study subjects. In addition, this study investigated a wide range of micronutrients. Although we did not find total energy to be a significant confounder in our study, we controlled for energy intake in the analysis of nutrients in order to adjust for potential general over- or under-reporting of all foods.
In addition to limitations inherent in case–control studies of past diet, other limitations of this study deserve mention. The dietary questionnaire was limited to fifty-five items, was not validated, nor had reliability measured, and portion sizes were not asked. The questions about vegetables and fruits were limited and did not ascertain various forms of cooked preparation, despite asking about consumption of ‘raw’ vegetables. The questionnaire asked about past diet, and responses may have been subject to recall bias. When differences in dietary recall occur non-differentially with respect to case–control status, estimates of risk are typically biased towards the null. If recall is differential, then risk estimates could be biased in either direction. It is known that although diet has some consistency over time, reported food intakes may not accurately reflect past behaviour(Reference Dwyer, Gardner and Halvorsen58). Dietary changes may also have occurred in the food supply (marketplace) over the past 20 years. Survey data suggest that the amount and proportion of energy from total fat and saturated fat have steadily declined over the last 20 years in the USA. Little is known about changes in fruit and vegetables intake although carbohydrate intake has increased(Reference Kant and Graubard59). Given that 99 % of the participants in our study were Whites, the present results may have limited generalisability to other racial/ethnic groups. Some observed associations may have been due to chance.
While RCC is not common in the general population, it is increasing, both in the USA and worldwide, despite a decrease in smoking rates in affected populations. It would therefore be worthwhile to further evaluate these findings in larger representative prospective studies, especially in older, non-smoking populations.
Acknowledgements
The present research was supported by the Intramural Research Program of the National Institutes of Health, National Cancer Institute, Division of Cancer Epidemiology and Genetics and Sydney University, NSW, Australia Sabbatical Program for K. E. B. In addition, we acknowledge the invaluable support of David Check, research assistant, Biostatistics Branch, Division of Cancer Epidemiology and Genetics, National Cancer Institute, National Institutes of Health. Contributions of the co-authors were as follows: K. P. C. and C. F. L. designed the study and had overall responsibility for the project; A. G. E. designed the collection of dietary information; C. F. L. was responsible for overseeing the subject selection and data collection; G. G., L. K., B. I. G., K. E. B. and B. C.-H. C. conducted the data analysis; K. E. B., L. K. and B. I. G. drafted the paper; and all authors contributed to the final completion of the manuscript. None of the authors had any conflicts of interest (personal, commercial, political, academic or financial).
