Intake of vegetables, fruits, carotenoids and vitamins C and E and pancreatic cancer risk in The Netherlands Cohort Study
Article first published online: 27 APR 2011
Copyright © 2011 UICC
International Journal of Cancer
Volume 130, Issue 1, pages 147–158, 1 January 2012
How to Cite
Heinen, M. M., Verhage, B. A.J., Goldbohm, R. A. and van den Brandt, P. A. (2012), Intake of vegetables, fruits, carotenoids and vitamins C and E and pancreatic cancer risk in The Netherlands Cohort Study. Int. J. Cancer, 130: 147–158. doi: 10.1002/ijc.25989
- Issue published online: 27 OCT 2011
- Article first published online: 27 APR 2011
- Accepted manuscript online: 15 FEB 2011 02:33PM EST
- Manuscript Accepted: 26 JAN 2011
- Manuscript Received: 13 SEP 2010
- pancreatic cancer;
- cohort study;
- microscopic confirmation
Epidemiological data investigating the relation between fruit and vegetable consumption and pancreatic cancer risk have shown inconsistent results so far. Most case-control studies observed an inverse association with total fruit and vegetable consumption, whereas results from most cohort studies have largely been null. We examined prospectively the relation between pancreatic cancer risk and intake of vegetables, fruits, carotenoids and vitamins C and E. The Netherlands Cohort Study consisted of 120,852 men and women who completed a questionnaire at baseline in 1986, including a validated 150-item food-frequency questionnaire. After 16.3 years of follow-up, 423 cases were available for analysis. Total vegetable and total fruit consumption were not associated with pancreatic cancer risk (highest vs. lowest quintile, multivariable-adjusted hazard rate ratio = 1.23, 95% confidence interval: 0.86-1.75 and multivariable-adjusted hazard rate ratio = 0.90, 95% confidence interval: 0.66-1.24, respectively). Also, for cooked vegetables, raw vegetables and vegetables and fruits classified into subgroups, no associations were observed. Dietary carotenoids, vitamin C and E intake and supplements containing vitamin C or E were not associated with pancreatic cancer risk. The results were not modified by sex, smoking status and body mass index. In conclusion, we observed no association between a high consumption of vegetables and fruits and pancreatic cancer risk in this large cohort study, which is in agreement with previous prospective studies. Furthermore, we observed no association between the intake of carotenoids, vitamins and vitamin supplements and pancreatic cancer risk.
Pancreatic cancer is the 5th leading cause of death in Europe and 4th in the United States.1, 2 Pancreatic cancer is diagnosed most often at advanced stages and patients diagnosed with pancreatic cancer have a 5-year survival rate of 6% or less.2, 3 So far, cigarette smoking, diabetes mellitus and body fatness are identified as risk factors.4–6
Fruits and vegetables contain numerous substances with potential anticarcinogenic activity (including vitamins, carotenoids and Allium compounds)7 and could therefore play a role in prevention of pancreatic cancer. Potential mechanisms of action include antioxidant protection against free-radical damage to DNA, enhancing immune function and inhibiting insulin-like growth factor (IGF) binding to IGF-receptors.4, 7 In addition, short-term animal experiments suggest that beta-carotene and the vitamins C and E hinder the development of preneoplastic lesions in both rat and hamster pancreas,8 but long-term studies demonstrated this inhibiting effect only for beta-carotene and vitamin C and only in rat pancreas.8, 9
Epidemiological data have shown inconsistent results so far. Most case-control studies have observed an inverse association with total fruit and vegetable consumption.10–15 Among specific subgroups of vegetables, the most consistent association has been found for cruciferous or Brassica vegetables.11, 12, 15 On the other hand, results from most cohort studies have largely been null.16–22 The results from the Multiethnic Cohort Study suggests that vegetables may afford some protection against pancreatic cancer in high-risk subgroups, namely current smokers and overweight/obese persons.18 Regarding the relation between antioxidant intake and pancreatic cancer risk, intake of vitamin C and beta-carotene have been investigated most often, showing both inverse associations11, 14, 23 and no association.15, 19, 20, 24, 25 Data on use of vitamin supplements has been very sparse;10, 15, 20, 24 of these studies only one observed an inverse association with vitamin C supplement use.15 A recent Expert Panel Report found only suggestive evidence that fruits protect against pancreatic cancer, whereas the evidence was inconclusive for vegetables and vitamin C.4
In our study, we investigated the association between pancreatic cancer risk and the overall consumption of vegetables and fruits and consumption of subgroups of vegetables and fruits in a large prospective cohort study in The Netherlands. In addition, we investigated the relation between pancreatic cancer risk and dietary carotenoids, vitamins C and E and supplements containing vitamin C and E.
Material and Methods
Study population and cancer follow-up
The study design of The Netherlands Cohort Study (NLCS) has been reported in detail elsewhere.26 Briefly, the NLCS was initiated in September 1986 and included initially 58,279 men and 62,573 women aged 55-69 years from 204 Dutch municipalities with computerized population registries. A self-administered food frequency and lifestyle questionnaire was completed at baseline. For efficiency in the processing of the questionnaire and follow-up, the case-cohort approach was used.27 Incident cases were derived from the entire cohort, whereas the person-years at risk were estimated from a random sample of 5,000 subjects (2,411 men and 2,589 women). This subcohort was chosen immediately after baseline and followed up for vital status information. The entire cohort is being monitored for cancer occurrence by annual record linkage to The Netherlands Cancer Registry and The Netherlands Pathology Registry.28, 29 A total of 16.3 years of follow-up (baseline to December 2002) was used for the current analysis. Only one subcohort member was lost to follow-up and completeness of follow-up was estimated to be >96%.30
All prevalent cancer cases at baseline other than skin cancer were excluded, resulting in a subcohort of 4,774 men and women. Of the 567 incident pancreatic cancer cases (ICD-O-3 code C25), cases with endocrine subtypes (ICD-O-3 code C25.4; n = 1) were excluded. Sixty two percent of the 566 pancreatic cancer cases were microscopically confirmed pancreatic cancer (MCPC, n = 350), whereas such confirmation was lacking for 38% (n = 216). Diagnosis of the latter group was made by the treating clinician and was based on clinical symptoms, physical examination and imaging results and data were abstracted and recorded by a trained tumor registrar.31 The NLCS has been approved by the institutional review boards of the TNO Nutrition and Food Research Institute (Zeist, The Netherlands) and Maastricht University (Maastricht, The Netherlands).
The dietary section of the baseline questionnaire was a 150-item semi-quantitative food-frequency questionnaire (FFQ), which concentrated on habitual consumption during the year preceding the start of the study. Questionnaire data were key-entered and processed for all incident cases in the cohort and subcohort members in a standardized manner, blinded with respect to case/subcohort status. This was done to minimize observer bias in the coding and interpretation of the data.
Data were obtained concerning consumption frequency of vegetables (i.e., Brussels sprouts, cauliflower, cabbage [white/green], kale, string beans, broad beans, spinach, endive [raw and cooked], lettuce, carrots [raw and cooked], sweet peppers, sauerkraut, tomatoes, red beets, mushrooms, gherkins, rhubarb, leek and onions), for summer and winter separately and fruits (i.e., mandarins, oranges, grapefruits, orange/grapefruit juice, grapes, bananas, apples/pears and strawberries). Based on data from the Dutch Nutrition Survey,32 onions and sweet peppers were considered to be eaten cooked, whereas tomatoes were considered to be eaten raw. The questionnaire covered most vegetables and fruits eaten regularly in 1986, with the exception of chicory, red cabbage and cucumber. In The Netherlands, broccoli was a rarely available vegetable in 1986 and therefore not included. Furthermore, in an open-ended question, participants could enter which other foods they consumed on a regular basis as well as the frequency (number of times per week) and amount of consumption on each occasion.
Consumption frequency was specified by using categories ranging from “never or less than once per month” to “three to seven times per week” for vegetable consumption and to “six or seven times per week” for fruit consumption. In addition, for individual fruit items, the amount consumed on each consumption day was asked. For onions and tomatoes, participants were asked to report the number they usually ate per week; for sweet peppers per month; and for mushrooms, how many 250-g boxes per month. Frequency of consumption and usual serving size of tomato/vegetable juice, processed orange/grapefruit juice and other fruit juices was also asked.
Participants were asked about usual serving sizes for string beans and cooked endive only; the mean of these serving sizes were used as a representative of solid and leafy vegetables, respectively. This average individual serving size was multiplied with a vegetable-specific factor calculated from pilot study data, to derive an individual serving size for each vegetable. This procedure was chosen because in a pilot study it was shown that serving sizes of different types of cooked vegetables were correlated within subjects. Mean daily vegetable consumption (grams/day) was calculated by multiplying the frequency of consumption and serving size. Frequency of intake and standard serving sizes were used to calculate consumption of individual fruit items in grams/day.
The mean daily intake of vitamins C and E were calculated by using the computerized Dutch food composition table.33 For calculating the intake of specific carotenoids, an additional food composition table was used. Briefly, foods that are the main sources of carotenoids (e.g., vegetables) were sampled and analyzed for alpha-carotene, beta-carotene, lutein, zeaxanthin and lycopene; the database was completed with data from the literature and information from food manufacturers.34 In the carotenoids food composition table, lutein and zeaxanthin were combined, because most literature sources had not distinguished these two carotenoids. Most vegetables, however, contain primarily lutein and only minor amounts of zeaxanthin. Information on dietary supplement use was collected with an open-ended question with space for four different supplements.35 Participants were asked whether they used vitamin tablets, drops, or other supplements during the 5 years before baseline. Furthermore, they were asked what type of supplement they have used, type of brand, what dosage and for how long they have used the supplement.
Subjects with incomplete or inconsistent dietary data (336 subcohort members, 46 cases) were excluded from analyses.36 Throughout the FFQ data cleaning was conducted using standardized algorithms to detect and, in some cases, correct likely errors, while tallying the errors for each person. Questions on vegetable consumption appeared early in the food frequency questionnaire. This led to some subjects' making mistakes on these particular items (e.g., improbably high summed frequencies for vegetable consumption, errors in separate consumption frequencies for summer and winter and improbably high or low reported portion sizes), while items on other food groups appearing later in the questionnaire were filled out without any problems. When more than three errors were encountered on the vegetable items, that subject was excluded from the analyses of vegetable consumption (241 subcohort members, 26 cases). The FFQ had been validated and tested for reproducibility.36, 37 Crude Pearson correlation coefficients between the 9-day diet record and the questionnaire were 0.74 for energy and 0.58 for vitamin C. The Spearman correlation coefficients for total vegetables and total fruits were 0.38 and 0.60, respectively.36 On average, vegetable consumption appeared to be slightly overestimated and fruit consumption to be underestimated by the FFQ as compared to the diet records.36
Analyses were performed for total vegetable consumption, total fruit consumption, consumption of vegetables and fruits combined, cooked and raw vegetables, vegetables categorized in subgroups (Brassica vegetables, cooked and raw leafy vegetables, Allium vegetables and legumes), total fruit, citrus fruit and consumption of individual vegetables and fruits as listed in the questionnaire. “Total vegetable consumption” is the summed total for all vegetables mentioned in the questionnaire and in the open-ended question, excluding dried pulses. Dried pulses were considered only in the analysis of legumes. The composition of each vegetable and fruit group is given in the “Appendix” section. In addition, analyses were performed for tomato/vegetable juice, processed orange/grapefruit juice and other juices. Other exposure variables were the carotenoids alpha-carotene, beta-carotene, lutein + zeaxanthin, lycopene and beta-cryptoxanthin, the vitamins C and E and use of any supplement—including multivitamins—containing vitamin C or E.
Participants were categorized according to quintile of intake of relevant food groups or nutrients (with the lowest quintile of intake regarded as the reference group), depending on the sex-specific distribution in the subcohort. For vitamin C, however, the validation study had pointed out that quintiles 2 and 3 and quintiles 4 and 5 could not be distinguished; therefore, we reduced vitamin C intake to three categories.36 Participants were categorized as users or nonusers of supplements containing vitamin C or E. Continuous variables were constructed as well. For vegetables and fruits an increment of 25 g/day was used based on data of the pilot study. This increment corresponds to a consumption frequency of approximately once per week for cooked vegetables.
Age- and sex-adjusted and multivariable-adjusted hazard rate ratios (HRs) and corresponding 95% confidence intervals (95% CIs) were estimated by using Cox proportional hazards models. The total person-years at risk, estimated from the subcohort, were used in the analyses.38 Standard errors were estimated by using a robust covariance matrix estimator to account for increased variance due to sampling from the cohort.39 All analyses were conducted for both sexes combined and separately for men and women.
Based on literature, the following variables were considered as potential confounders: age, sex, smoking, body mass index (BMI), intake of energy, coffee and alcohol, total red meat consumption, level of education, nonoccupational physical activity, family history of pancreatic cancer, history of diabetes mellitus, hypertension, gall-stones, cholecystectomy and gastric ulcer. These potential confounding variables were added to the multivariable-adjusted model if they (i) were associated with the disease and with total vegetable and total fruit intake and (ii) changed the age- and sex-adjusted regression coefficients by at least 10 percent (using a backwards stepwise procedure). This resulted in a multivariable-adjusted model including age at baseline (years), sex, cigarette smoking (current smoking: yes/no; number of cigarettes smoked per day; number of years of smoking), BMI (kg/m2), intake of energy (kcal/day), coffee (number of cups/day) and alcohol (g/day), total red meat consumption (g/day), family history of pancreatic cancer (yes/no) and history of diabetes mellitus (yes/no). The independent contribution of each vegetable subgroup was assessed by an analysis in which all vegetable subgroups were included in the model simultaneously. For analysis on antioxidant intake, the independent contribution of each specific vitamin and carotenoid to the risk of pancreatic cancer was assessed by an analysis in which all these were included in the model simultaneously. In additional analyses, the HRs were adjusted for total vegetable and fruit intake. For each analysis, trends were evaluated with the Wald test by assigning participants the median value for each level of the categorical exposure variable among the subcohort members and this variable was entered as a continuous term in the Cox regression model.
To permit comparison, we restricted age-adjusted analyses to subjects included in multivariable-adjusted analyses (e.g., with no missing values on confounding variables), which left 3,937 subcohort members (1,930 men and 2,007 women) and 448 exocrine pancreatic cancer cases (240 men and 208 women) for analysis on fruit consumption and on intake of carotenoids, vitamins and vitamin supplements. For the analysis on vegetable consumption, 3,734 subcohort members and 428 exocrine pancreatic cancer cases were available. The proportional hazards assumption, which was tested using the scaled Schoenfeld residuals,40 was violated for many of the exposure variables. Because early symptoms of disease before diagnosis could have influenced the results, the early cases (diagnosed within 2 years after baseline) were excluded; this resolved our problem of assumptions being violated. Therefore, all analyses were done excluding the first 2 years of follow-up: 69 subcohort members and 25 pancreatic cancer cases were excluded for the analysis on fruits, juices, carotenoids and vitamin intake and use of vitamin supplements. For the analysis on vegetable intake, 64 subcohort members and 22 pancreatic cancer cases were excluded.
In our study, the overall analyses were performed on all pancreatic cancer cases. We restricted additional analyses to MCPC cases to create a group with a higher degree of diagnostic certainty of pancreatic cancer, which was shown to be important in previous studies.41, 42 We also conducted analyses separately for current, former and never smokers to determine whether smoking modifies the association of total vegetable and total fruit intake with risk of pancreatic cancer. Also, analyses for total vegetable and fruit intake were conducted stratified by BMI level (cutoff: 25 kg/m2) to examine whether fruit and vegetable intake may differentially affect those at higher risk. In addition, interactions on a multiplicative scale of total vegetable and total fruit consumption with smoking status and BMI were tested. All analyses were performed using the STATA statistical software package (intercooled STATA, version 9; Stata Corporation, College Station, TX). All p values were based on two-sided tests and considered statistically significant if <0.05.
In Table 1, baseline characteristics are presented. Most characteristics did not differ between pancreatic cancer cases and subcohort members; however, there were more current smokers and more subjects with a family history of pancreatic cancer among pancreatic cancer cases than among subcohort members. In Table 2, we present baseline dietary intakes of the specific vegetables and fruits as listed in the questionnaire. Because the specific vegetables and fruits had a right-skewed distribution, we present the median and interquartile range. The specific vegetables and fruits are presented in increasing order of the percentage of nonusers. The most eaten vegetables in our population were string beans, cauliflower and lettuce, whereas gherkins and raw carrots were only consumed by ∼30% of our participants. Apples, pears, strawberries and oranges were the most consumed fruits, whereas grapefruits, raisins and other dried fruit were the least consumed fruits in our population. No real differences were observed between cases and subcohort members regarding the intake of specific vegetables and fruits.
In Table 3, HRs are presented for quintiles of total vegetable and fruit intake, total vegetable intake, intake of vegetable subgroups, total fruit intake and citrus fruit intake. After adjustment for age and sex, trends in risk for cooked vegetables and Brassica vegetables were (statistically significantly) positive; however, none of the HRs in the categorical analyses were significant (Table 3). When the confounders were included in the model, these positive trends became nonsignificant. For all other vegetable and fruit groups, we observed no associations. All HRs for an increment in daily mean intake of 25 g/day for vegetables and fruits, were ∼1 (Table 3). When we included simultaneously all vegetable subgroups (continuously) in the model, results were similar (results not shown). Additional adjustment for fruit and vegetable consumption did also not alter the results (results not shown). We have also investigated whether specific vegetables and fruits were associated with pancreatic cancer risk; no associations were observed (results not shown). Also the intake of tomato/vegetable juice, processed orange/grapefruit juice and other fruit juices were not associated with pancreatic cancer risk (results not shown).
In Table 4, HRs are presented for carotenoids and vitamin intake. We did not observe any association for the intake of carotenoids, vitamin C and E and intake of supplements containing vitamin C or E. We simultaneously included the carotenoids and vitamins in a model; this did not alter the HRs (data not shown). Also, additional adjustment for fruit and vegetable consumption did not alter the results (results not shown).
Results did not differ between men and women and using the total follow-up period including the first 2 years of follow-up (results not shown). The exclusion of participants without histological confirmed pancreatic cancer did not change the null findings (results not shown). We investigated whether smoking status (never/ex/current) and BMI (<25 kg/m2vs. ≥25 kg/m2) modified the associations between vegetable and fruit intake and pancreatic cancer risk; risk estimates were not different for never, former and current smokers and for normal versus overweight persons (results not shown). In addition, the multiplicative interaction terms were not statistically significant (p for interaction = 0.60 for vegetables and 0.79 for fruit).
In our study no association was observed between pancreatic cancer risk and consumption of vegetables, fruits and juices. We also observed no association between pancreatic cancer risk and the intake of carotenoids, vitamins and vitamin supplements. These results are in agreement with other cohort studies. Results were not modified by sex, smoking status and BMI.
Inverse associations have been observed with vegetable intake in several case-control studies. The vegetable items reported in these studies were diverse, including total vegetable plus fruit,12, 13 total vegetables,11, 12, 15Brassica vegetables,11, 13, 15 dark green leafy vegetables,12Allium vegetables,12 carrots,12, 13 and raw vegetables.11, 13, 15 Also for fruit items inverse associations have been reported for several types of fruit, including total fruit,10, 11, 13 citrus fruit,13 oranges,14 and bananas.14 Cohort studies on the other hand, have mainly reported null associations.16–22 Only a few prospective studies observed protective effects of vegetables and fruits on pancreatic cancer, showing inverse associations for total fruit in Japanese men (not in women),43 cabbage consumption among Swedish women,44 vegetarian protein products, beans, lentils and peas and dried fruit intake among Adventists.45 However, two of these studies had low case numbers (<150).44, 45 The Multiethnic Cohort Study observed an increased pancreatic cancer risk for high intake of fruit; this was, however, most apparent among nonsmokers and no association has been found for citrus fruit.46 For fruit juices mixed results have been observed as well, showing inverse associations12, 43 and no associations.11, 21
The Multiethnic Cohort Study found no association between total vegetable intake and pancreatic cancer risk overall; they found, however, some protection against pancreatic cancer in high-risk subgroups (i.e., current smokers and overweight/obese persons). They observed a significantly inverse association with dark green vegetable consumption among current smokers.18 They also observed an inverse association with total vegetables in overweight/obese persons (≥25 kg/m2). We, as well as others,21, 47 did not observe such findings.
For carotenoids and vitamins, case-control studies observed inverse associations for beta-carotene,11, 23 lycopene (in men),25 vitamin C,11, 14, 23 and E;14, 23 cohort studies reported only null findings on these carotenoids and vitamin intake.19, 20 Other carotenoids, such as alpha-carotene, lutein plus zeaxanthin and beta-cryptoxanthin, have only been investigated in case-control studies so far,15, 25 showing no associations with pancreatic cancer risk.
Inconsistencies could have occurred because case-control studies are prone to more biases compared to cohort studies, including recall bias; risk estimates might be either exaggerated or underestimated because dietary intake is assessed in cases after diagnosis. Also selection bias is a problem due to high and rapid fatality rates of pancreatic cancer cases. Differential misclassification of the exposure could also have occurred due to the need to use next-of-kin interviews in case-control studies because pancreatic cancer is rapidly fatal. In addition, for several studies no dietary information was available on individual vegetables or fruits16, 17, 22 and some studies had small sample sizes (n cases <150).19, 22, 44, 45
In our study, an extensive list of vegetables and fruits was assessed. On the other hand, dietary assessment is liable to error and may have resulted in misclassification of exposure. Vegetables are generally considered as food items that are not very easy to assess in FFQs, particularly if portion sizes have to be estimated. We have intended to minimize the amount of uninformative data. Subjects with incomplete or inconsistent dietary data and, specifically, those participants who appeared not to have understood how to answer the questions on vegetable consumption, were excluded. If misclassification has occurred, we expect this to be nondifferential and risk estimates will be most likely biased towards the null value. In the NLCS validation study, the correlation coefficient between the 9-day diet record and the FFQ for total vegetable consumption was 0.38.36 This correlation is low, but comparable to the figure reported for other prospective studies.43, 44, 48 One of the reasons for the low correlation could be that our study population may have been too homogeneous regarding intake and therefore may have yielded too little contrast between highest and lowest quintile of total vegetable consumption to detect differences in pancreatic cancer risk. Because of individual preferences, however, contrast in consumption frequency of many specific vegetables as well as for fruit is much higher. Although we cannot entirely exclude the possibility that the absence of protective effects of vegetables and fruits on pancreatic cancer is due to measurement error or too little contrast in our data, this can not be unique for our study and it does not explain why especially case-control studies observed protective associations. A large European cohort study—which also did not observe an association between pancreatic cancer risk and fruit and vegetable consumption—has a wider range of fruit and vegetable intake compared to other prospective cohort studies, caused by inclusion of participants from Northern to Southern Europeans countries.21 However, their range of fruit and vegetable intake was comparable to ours.
Strengths of our study include the possibility to further restrict the analyses to MCPC cases only42 and the large sample size. Differential follow-up is unlikely to have made a material contribution to our findings, as completeness of follow-up was high.30 The prospective design avoided recall bias and the need to use next-of-kin respondents.
In conclusion, we observed no association between pancreatic cancer risk and a high consumption of vegetables and fruits in the NLCS, which is in agreement with previous prospective studies. Furthermore, we observed no association between the intake of alpha-carotene, beta-carotene, lutein plus zeaxanthin, beta-cryptoxanthin, lycopene and vitamins C and E and pancreatic cancer risk.
The authors are indebted to the participants of this study and further wish to thank the cancer registries (IKA, IKL, IKMN, IKN, IKO, IKR, IKST, IKW, IKZ and VIKC) and The Netherlands nationwide registry of pathology (PALGA). They also thank Dr. A. Volovics and Dr. A. Kester for statistical advice; Dr. L. Schouten, Ms. S. van de Crommert, Ms. H. Brants, Ms. J. Nelissen, Ms. C. de Zwart, Ms. M. Moll and Ms. A. Pisters for assistance and Mr. H. van Montfort, Mr. T. van Moergastel, Ms. L. van den Bosch, Mr. R. Schmeitz for programing assistance.
- 4World Cancer Research Fund/American Institute for Cancer research. Food, nutrition, physical activity, and the prevention of cancer: a global perspective. Washington DC: AICR, 2007.
- 30Estimation of the coverage of Dutch municipalities by cancer registries and PALGA based on hospital discharge data. Tijdschr Soc Gezondheidsz 1994; 72: 80–84., , .
- 32Anonymous. What does The Netherlands consume; results of the national food consumption survey of 1987–1988 [in Dutch]. Rijswijk, The Netherlands: Distributiecentrum DOP, Ministry of Health, Welfare and Culture and Ministry of Agriculture and Fisheries, 1988.
- 33Anonymous. NEVO-tabel: Nederlands voedingsstoffenbestand 1986–1987. The Hague, The Netherlands: Voorlichtingsbureau voor de voeding, 1986; (Nevo table: Dutch Food Composition Table 1986–1987 [in Dutch].
- 38Methods for the analysis of case-cohort studies. Biomed J 1997; 39: 195–214., .