| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
ORIGINAL RESEARCH COMMUNICATION |
1 From the Department of Hygiene, Medical School, University of Catanzaro "Magna Græcia", Catanzaro, Italy (MP, CP, and CGAN) and the Department of Public, Clinical and Preventive Medicine, Medical School, Second University of Naples, Naples, Italy (IFA).
2 Address reprint requests to IF Angelillo, Department of Public, Clinical and Preventive Medicine, Second University of Naples, Via Luciano Armanni 5, 80138 Naples, Italy. E-mail: italof.angelillo{at}unina2.it.
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONSUBJECTS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
Objective: A meta-analysis was performed to arrive at quantitative conclusions about the contribution of fruit and vegetable intakes to the occurrence of oral cancer.
Design: A comprehensive, systematic bibliographic search of medical literature published up to September 2005 was conducted to identify relevant studies. Separate meta-analyses were conducted for fruit and vegetable consumption. The effect of portion or daily intake of fruit or vegetables on the risk of oral cancer was calculated. A multivariate meta-regression analysis was performed to explore heterogeneity. This multivariate meta-regression analysis examined the effect of quality score, the type of cancers included, citrus fruit and green vegetable consumption, and the time interval for dietary recall of the studies on the role of fruit or vegetable consumption in the risk of oral cancer. The presence of publication bias was assessed with a funnel plot for asymmetry.
Results: Sixteen studies (15 case-control studies and 1 cohort study) met the inclusion criteria and were included in the meta-analysis. The combined adjusted odds ratio (OR) estimates showed that each portion of fruit consumed per day significantly reduced the risk of oral cancer by 49% (OR: 0.51; 95% CI: 0.40, 0.65). For vegetable consumption, the meta-analysis showed a significant reduction in the overall risk of oral cancer of 50% (OR: 0.50; 95% CI: 0.38, 0.65). The multivariate meta-regression showed that the lower risk of oral cancer associated with fruit consumption was significantly influenced by the type of fruit consumed and by the time interval of dietary recall.
Conclusion: The consumption of fruit and vegetables is associated with a reduced risk of oral cancer.
Key Words: Fruit • meta-analysis • oral cancer • vegetables
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONSUBJECTS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
In epidemiologic studies, the primary risk factors for oral cancer have been well documented and include past and present betel-quid chewing, tobacco smoking, and consumption of alcohol in the form of beer, wine, or liquor. Poor dietary practices and nutritional deficiencies have been linked to a risk of developing oral cancer, but recent reports have examined the effect of fruit and vegetable intake, which contain essential vitamins and other nutrients, on the incidence of oral cancer. Findings from these reports, however, have been mixed; some observational studies have shown a protective effect with the consumption of fruit and vegetables, whereas others have not shown this benefit. Many unanswered questions remain. For example, there is no consensus on whether the benefit observed with fruit and vegetable consumption is observed with all fruit and vegetable products, or whether this benefit varies with different types and preparations (ie, cooked compared with raw) of fruit and vegetables. A proven association between the risk of oral cancer and the consumption of fruit and vegetables would have considerable public health and nutritional implications. Our study provided a unique opportunity to answer some of these questions, because we performed a meta-analysis that aimed to use all relevant published literature from observational studies to arrive at a quantitative conclusion about the contribution of fruit and vegetable intakes to the occurrence of oral cancer.
|
SUBJECTS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONSUBJECTS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
Inclusion criteria
Two investigators independently reviewed all potentially relevant articles to determine whether an article met the general inclusion criteria, and disagreement was resolved by discussion between the investigators. Studies were included in the meta-analysis if they met all of the following criteria: 1) had original data from case-control or cohort studies; 2) the primary outcome was clearly defined as at least some of the cancers of the mouth, pharynx, and hypopharynx [as defined by codes 141, 143–146, 148, and 149 of the ninth revision of the International Classification of Diseases (ICD-9; 2) or other classifications that included these sites]; 3) the exposure of interest was measured as servings (per day, per week, or per month) of fruit, vegetables, or both (other exposure measurements, such as grams per day, were excluded unless data were provided to transform information to servings per day); 4) provided relative risk (RR) estimates and their 95% CIs or sufficient data to calculate these numbers; 5) were English language studies; and 6) were published up to September 2005. If a study appeared in more than one article, data from the most recent publication were used for the statistical analysis. Studies restricted to oral cancer in subjects aged <45 y were excluded.
Assessment of study quality
Two investigators independently reviewed the studies included in the meta-analysis to arrive at an assessment of the quality of the individual study. Each article was read and scored for quality, and investigators, institutions, country, and journal were blinded for each article. Because no validated tools exist for a quality assessment of outcome studies, we developed a criteria list that incorporated elements of previously published criteria (3, 4) for the assessment of quality items in epidemiologic studies. The investigators discussed their evaluation; discrepancies were resolved through discussion and rereading. The list was composed of items felt to be important for the quality of each observational study, including the study design [selection bias; score ranging from 0 (worst) to 6 (best)], the adjustment of confounding variables (score ranging from 0 to 16, worst to best), the exposure assessment (misclassification bias; score ranging from 0 to 5, worst to best), and the data analysis (score ranging from 0 to 2.5, worst to best). Each subscore was calculated as the percentage of applicable quality criteria that were met in each study; therefore, each subscore for a study could range from 0% (none of the quality criterion was met) to 100% (all the quality criteria were met). The cumulative quality score was a weighted average of the 4 percentages. To avoid selection bias, no study was rejected because of these quality criteria.
Data extraction
All data from the studies were independently reviewed and extracted with a standardized data-collection form by 2 investigators. Differences between reviewers were resolved by discussion and, when necessary, through consultation. The following characteristics were recorded from each study: 1) the first author's name, year of publication, and country of the population; 2) the study design; 3) the classification used for the disease; 4) the number and ages of the subjects; 5) any confounding factors for matching or adjustment; 6) the methods used for collection of data on exposure; and 7) the odds ratio (OR) of oral cancer associated with fruit or vegetable consumption and the corresponding 95% CI in each subgroup. For a meta-analysis on the relation between diet and the risk of oral cancer, we selected all observational studies that related either vegetable or fruit intake to the incidence, mortality, or prevalence of oral cancer. Furthermore, a retrospective assessment of the diet was feasible because oral cancer is not immediately fatal. The studies included in our meta-analysis often differed in the measurement of fruit and vegetable consumption, with many different categories used to indicate the extent of consumption. All studies were stratified by fruit or vegetable intakes to evaluate any dose-response relations, and some of the studies performed multivariate analyses to adjust for several confounders. For the published results of each of the selected studies, data were extracted to permit the calculation of both unadjusted and adjusted ORs with 95% CIs to estimate the association between fruit and vegetable consumption and the risk of oral cancer.
Meta-analyses
Separate meta-analyses were conducted for fruit and vegetable consumption. The effect of portions or daily intakes of fruit or vegetables consumed on the risk of oral cancer was calculated. The analysis was repeated with the results for citrus fruit and green vegetable consumption obtained from studies in which these results were available. For each study, a weighted log-linear regression analysis of the adjusted ORs was performed according to the mean, median, or midpoint of fruit or vegetable consumption, except in the case of studies with only 2 exposure categories in which the value of the logarithm of the OR for one portion was used. In open-ended categories, the number of portions was chosen proportionally to those of the other categories. Weights were calculated according to the methods described by Greenland (5) and Berlin et al (6). Combined risk estimates were calculated by using the risk estimates that reflected the greatest degree of control for other environmental and dietary risk factors (OR adjusted for confounding factors). Potential sources of heterogeneity between the studies were examined by using the method developed by DerSimonian and Laird (7), which calculates the between-study variation based on the Q statistic. We considered that there was statistically significant heterogeneity when the P value between the results of the included studies was below 0.1. In cases with heterogeneity, we applied random-effects models as opposed to fixed-effect models because the former include both within-study sampling error (variance) and between-study variation in the assessment of the uncertainty (95% CI) of the results of a meta-analysis.
Sensitivity analyses
To explore the reasons for the observed heterogeneity, sensitivity analyses were performed by grouping studies that showed more similar characteristics, such as similar cases according to ICD-9 codes, cases restricted to exposure to citrus fruit and green vegetables, those that presented disaggregated data by sex, or those that were adjusted for a core of variables, such as age, sex, cigarette smoking, and alcohol consumption. Finally, we investigated the effect of the poor-quality studies on the overall effect size by performing a sensitivity analysis on the results by 2 subgroups, which were based on individual scores above or below the median.
To additionally explore heterogeneity, a multivariate meta-regression analysis was performed. This multivariate meta-regression analysis examined the effect of certain variables, such as quality score, type of cancers included, citrus fruit and green vegetable consumption, population studied (men, women, or both) and time interval for dietary recall, on the role of fruit or vegetable consumption in the risk of oral cancer. The random-effects regression model relates the treatment effect to the covariates, assuming a normal distribution for the residual errors with a within-study and an additive between-studies component of variance. The between-studies variance was estimated with an iterative procedure by using an estimate that is based on a restricted maximum likelihood method (8).
Finally, the presence of publication bias was assessed with a funnel plot for asymmetry, which is a scatter plot of the individual studies that relates the magnitude of the treatment effect to a measure of its precision (9); for formal statistical testing we used an adjusted rank correlation test and a regression asymmetry test (10, 11). All analyses were performed with STATA statistical software version 8.1 (12).
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONSUBJECTS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
The summary characteristics of all studies included in the meta-analysis are described in Table 1
. The sample size of the 16 included studies (15 case-control studies and 1 cohort study) varied between 92 and 1192 for the case subjects and between 106 and 36527 for the control subjects. The studies were geographically heterogeneous: most of the study populations involved European samples (n = 7), 5 studies were conducted in the Americas, and 4 were conducted in East Asia. The study populations in 3 case-control studies consisted solely of men and one study consisted solely of women. The age of the participants varied from 18 to 91 y. To give an indication of the actual OR found in the studies, we also show the ORs for the group with the different exposure for each study. All selected studies for both fruit and vegetable consumption reported either crude or adjusted ORs <1.0, which suggested that fruit and vegetable intake had a protective effect on oral cancer incidence, although a statistically significant positive effect (P < 0.05) compared with the reference category was observed in only 17 of the 32 comparisons investigating fruit consumption and in 15 of the 28 comparisons investigating vegetable consumption. In all except one study, dietary habits were collected by an in-person interview, and recall of lifelong dietary habits was referred to in 8 studies, recall of 2-y dietary habits before interview was referred to in 2 studies, and recall of 1-y dietary habit before interview was referred to in 4 studies.
|
. The overall quality ratings of the studies varied from 0.39 to 0.75, with a median of 0.66. In all case-control studies, case subjects and control subjects were identified without knowledge of exposure status and control status was not associated with fruit or vegetable intakes; moreover, in the only cohort study, exposed and nonexposed subjects were identified without knowledge of disease status. Validation of disease diagnosis by histology or other gold standards was satisfied in 87.5% studies. With the exception of 2 studies, all used standard definitions, such as the ICD (8th or 9th revision), to define the cancer outcome. Most studies (75%) presented demographic data and 68.8% conducted a statistical analysis on such data. The extent of adjustment for potential confounding factors in the relation between oral cancer and diet varied considerably across the studies. All studies adjusted for age and sex and 93.8% adjusted for tobacco smoking, but only 12.5% adjusted for tobacco chewing habits and 75% for alcohol consumption. Only 12 studies simultaneously presented adjusted ORs for known risk factors for oral cancer, including age, sex, cigarette smoking, and alcohol consumption. Statistical analyses in these studies included multivariate analyses with adjustment for confounders, and 93.8% of the studies listed P values and CIs. No studies performed any power calculations.
|
; citrus fruit consumption was taken into account when available. Fruit consumption reduced the occurrence of oral cancer in all 16 studies, because all estimates of the single studies derived from the log-linear model were below 1.0, but a statistically significant protective effect was only found in 13 comparisons. The combined adjusted OR estimates based on 16 studies showed that each portion of fruit consumed per day had an overall statistically significant effect on reducing the risk of oral cancer by 49% (OR: 0.51; 95% CI: 0.40, 0.65). An almost similar effect was reported for vegetable consumption: a significant reduction in the risk of oral cancer was found in 10 out of 15 studies, whereas the combination of all studies uncovered an overall significant reduction in the risk of oral cancer of 50% (OR: 0.50; 95% CI: 0.38, 0.65), and, in this case, data on green vegetables were used when available (Figure 2). However, the Q statistic test of homogeneity found a statistically significant heterogeneity across the various studies; the results for fruit consumption (Q: 1754.54; df: 15; P < 0.001) and for vegetable consumption (Q: 193.68; df: 14; P < 0.001) were heterogeneous.
|
|
).
|
).
|
).
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONSUBJECTS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
The strengths of the present meta-analysis include the acceptable methodologic quality of the studies on which the analysis is based, as well as the considerable number of studies and subjects included. Possible limitations of the study include the heterogeneity between the studies, including studies in which outcomes were recorded with different intakes of fruit or vegetables measured as servings per day, per week, or per month. Defining the amount of fruit or vegetable consumption at baseline (cohort study) or during the reference period (case-control studies) possibly caused the heterogeneity between the included studies, because the follow-up period and the reference dates varied between the studies. Therefore, the results of the meta-analysis must be interpreted with caution. Because of potential additional heterogeneity in the populations, designs, and analyses of the various studies, we assumed that the true effect being estimated would vary between the studies, in addition to the usual sampling variation in the estimates (within studies). To account for both sources of variation, a random-effects meta-regression analysis was used to combine the results of the primary studies. The random-effect approach provides some allowance for heterogeneity in studies beyond sampling error. This does not necessarily rule out the effect of heterogeneity between the studies, but one can expect a very limited influence because of it. It should also be noted that the phrase "oral cancer" indicates a heterogeneous group of disorders, and that most of the studies use different classification systems to define it. If vegetable exposures contributed to only a specific subset of oral cancers, then failure to find causal associations in the various studies may be the result of their failure to identify and count the appropriate endpoint cancers. That possibility, which would reduce the statistical power of individual studies, was partially addressed by pooling results of the individual studies in our meta-analysis. We thought that the different time of exposure and classifications to define the cancers could explain the statistical heterogeneity that was found between the study results, but heterogeneity persisted after we performed relevant subgroup analyses. Although the studies included in our analysis were heterogeneous, the relation is largely consistent.
The assessment methods for fruit and vegetable consumption may also vary between the studies. The assessment is usually based on self-reported habits, and such data are subject to recall errors. For example, control subjects may be more likely to underreport their consumption, whereas case subjects, because of symptoms (aches, etc) related to eating, may have posed more attention to food and diet recall. However, in the present meta-analysis, most of the studies used in-person interviews, whereas only one study used a self-administered questionnaire. Moreover, our meta-analysis was based on observational studies, mainly case-control studies, and the possibility of selection bias, misclassification bias related to exposure, and failure to consider potential confounders cannot be ruled out; these biases can lead to very precise, but spurious, results in a meta-analysis. In this case, with regard to misclassification of exposure, recalls of diet may be influenced by disease status, whereas selection bias may be related to higher participation rates of patients with cancer than of control subjects; additionally, participating control subjects are likely to be more health conscious than are nonparticipating control subjects and thus consume more fruit and vegetables. Therefore, results are to be interpreted with caution, because this could have led to a spurious inverse association between fruit and vegetable consumption and the risk of oral cancer.
It has been argued that because meta-analyses of observational studies may produce very precise, but spurious, results, a statistical combination of these data should not be the prominent component. However, the present meta-analysis allowed for a careful examination of possible sources of heterogeneity to systematically examine the strengths and weaknesses of the accumulated evidence, to contribute to substantial improvement of methodologic quality of research, and to identify potential biases, data gaps, and suggestions for future research.
A statistically significant funnel plot asymmetry of the effects of vegetables consumption on risk of oral cancer was detected with the Egger et al (ie, weighted regression; 11) method, but not with the Begg and Mazumdar (ie, rank correlation; 10) method, and this is compatible with a greater statistical power of the regression test. This asymmetry, which was due to the fact that smaller studies showed larger benefits, may be caused either by small "positive" studies that are more likely to be published than those with "negative" results or by small studies that, having a lower quality than the large studies, tend to exaggerate the effects of vegetable consumption. However, despite the significant heterogeneity detected, the results of the meta-regression and of the stratified analysis clearly indicated that vegetable consumption had a protective effect against the risk of oral cancer.
In conclusion, on the basis of epidemiologic evidence, we found that the consumption of fruit and vegetables was associated with a reduced risk of oral cancer. Prospective studies, which are less prone to recall and selection bias, are needed to confirm this result. In the case of an established protective effect of fruit and vegetable consumption on the risk of oral cancer, the provision of an easily understandable measure of the association may create the basis for a new dietary pattern characterized by a high consumption of fruit and vegetables.
|
ACKNOWLEDGMENTS |
|---|
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONSUBJECTS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  S. Kapoor Chemoprevention of Head and Neck Cancers: Promising New Biochemical Prospects J. Clin. Oncol., May 10, 2008; 26(14): 2417 - 2418. [Full Text] [PDF]
|
|
|
|
 |
|
 M. Hashibe, P. Brennan, S. Benhamou, X. Castellsague, C. Chen, M. P. Curado, L. D. Maso, A. W. Daudt, E. Fabianova, V. Wunsch-Filho, et al. Alcohol Drinking in Never Users of Tobacco, Cigarette Smoking in Never Drinkers, and the Risk of Head and Neck Cancer: Pooled Analysis in the International Head and Neck Cancer Epidemiology Consortium J Natl Cancer Inst, May 16, 2007; 99(10): 777 - 789. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 T. Asakage, A. Yokoyama, T. Haneda, M. Yamazaki, M. Muto, T. Yokoyama, H. Kato, H. Igaki, T. Tsujinaka, Y. Kumagai, et al. Genetic polymorphisms of alcohol and aldehyde dehydrogenases, and drinking, smoking and diet in Japanese men with oral and pharyngeal squamous cell carcinoma Carcinogenesis, April 1, 2007; 28(4): 865 - 874. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
