Association between intake of antioxidants and pancreatic cancer risk: a meta-analysis.

Abstract We conducted a meta-analysis to systematically evaluate the association between antioxidants intake and pancreatic cancer risk. Relevant articles were retrieved from PUBMED and EMBASE databases and standard meta-analysis methods were applied. Finally a total of 18 studies were included. Comparing the highest with lowest categories, higher dietary intakes of selenium, vitamin C, vitamin E, β-carotene and β-cryptoxanthin were significantly associated with reduced pancreatic cancer risk (for selenium, pooled OR = 0.47, 95%CI 0.26–0.85; for vitamin C, pooled OR = 0.68, 95%CI 0.57–0.80; for vitamin E, pooled OR = 0.70, 95%CI 0.62–0.81; for β-carotene, pooled OR = 0.74, 95%CI 0.56–0.98; for β-cryptoxanthin, pooled OR = 0.70, 95%CI 0.56–0.88). Lycopene intake was marginally associated with pancreatic cancer risk (pooled OR = 0.85, 95%CI 0.73–1.00), while no significant association was observed for α-carotene, lutein and zeaxanthin. In summary, higher dietary intake of selenium, vitamin C, vitamin E, β-carotene and β-cryptoxanthin was inversely associated with pancreatic cancer risk.


Introduction
Pancreatic cancer remains the 13th most commonly diagnosed cancer worldwide, and is the eighth leading cause of cancer death (Siegel et al. 2013). Early diagnostic methods for pancreatic cancer are inefficient and the prognosis is poor, with one-year survival rate of about 25% and five-year survival rate of 4-5%, and it is estimated that pancreatic cancer leads to 227,000 deaths each year (Verdecchia et al. 2007). Thus, it is an important issue to identify risk factors for pancreatic cancer and to prevent it accordingly.
Many risk factors for pancreatic cancer have been identified, including family history, smoking and type 2 diabetes (Klein et al. 2004;Huxley et al. 2005). Diet might be involved in the aetiology of pancreatic cancer, for example, vegetable and fruit consumption may decrease the risk of pancreatic cancer (Wiseman 2008). One explanation is that vegetable and fruit are rich in antioxidants, such as vitamin C, vitamin E and carotenoids, which may prevent pancreatic cancer by inactivating free radicals and reducing oxidative DNA damage (McCullough & Giovannucci 2004). This is important since DNA damage caused by heredity, smoking or other factors plays a fundamental role in pancreatic cancer development. Several studies have investigated the association between intake of antioxidants (including selenium, vitamin C, vitamin E, a-carotene, b-carotene, b-cryptoxanthin, lycopene, lutein and zeaxanthin) and pancreatic cancer risk, while there is still no clear conclusion (Stolzenberg-Solomon et al. 2002Bravi et al. 2011;Heinen et al. 2012;Banim et al. 2013;Han et al. 2013;Jansen et al. 2013). Therefore, the aim of our study is to assess the current evidence on the association between antioxidant nutrients intake and pancreatic cancer risk.

Literature search and study selection
This meta-analysis was designed, conducted and reported in adherence to the PRISMA statement (Moher et al. 2009). The literature search was conducted up to December 2014 in the PUBMED and EMBASE databases without restriction. The following terms were used in the literature search: (''antioxidant'' or ''vitamin'' or ''vitamin C'' or ''vitamin E'' or ''carotene'' or ''tocopherol'' or ''carotenoid'' or ''lycopene'' or ''selenium'' or ''b-cryptoxanthin'' or ''lutein and zeaxanthin'') AND (''pancreatic cancer'' or ''pancreatic ductal adenocarcinoma'' or ''pancreatic adenocarcinoma''). References cited in the selected articles and relevant reviews were also manually searched for potential missing studies. For study selection, titles and abstracts of publications identified from the initial search were first scanned, and then full papers were carefully reviewed for potential eligible studies.
Studies would be included in this meta-analysis if they met all the following criteria: (i) study should be designed as cohort, case-control or cross-sectional study; (ii) the exposure of interest was dietary antioxidant intake, while the outcome of interest was pancreatic cancer; (iii) odds ratio (OR) or relative risk (RR) estimates with 95% confidence intervals (95%CI) were reported or could be calculated.

Data extraction
Data of each study was independently extracted by two reviewers, and discrepancies were resolved by discussion or a third investigator. The following information was extracted from each study: first author/publication year, study design, sample size, sex and age of participants, country of origin, antioxidant kind, variables adjusted for in the study, and OR (or RR) estimates with 95%CIs for the highest versus lowest categories of dietary antioxidant intake. ORs (RRs) reflecting the greatest degree of control for potential confounders were adopted in the pooled analysis. For duplicated data, only the most detailed or recent information was extracted. We adopted the widely used Newcastle-Ottawa Scale to assess the study quality (Stang 2010).

Statistical analysis
The extent of heterogeneity was evaluated using Q test (with p value <0.1, suggesting significant heterogeneity) and I 2 index with I 2 >50%, suggesting substantial heterogeneity. Summary ORs (RRs) and 95%CI were calculated using a random-effects model when the heterogeneity was significant, and a fixed-effects model was applied otherwise. The primary meta-analyses were conducted to evaluate the association between antioxidants intake and pancreatic cancer risk. Metaregression, sensitivity and subgroup analyses were used for exploring source of heterogeneity with the following variables: design, country of origin and sample size. Meta-regression analyses would be applied for the meta-analyses with 10 or more studies. Besides, subgroup analyses were also applied to assess whether these variables would modify the results of meta-analysis. To evaluate the publication bias risk, funnel plots were evaluated and Begg's and Egger's tests were applied. However, for pooled analyses with fewer than 10 studies, we did not further conduct statistical tests for funnel plot asymmetry because of the limited test power (Sterne et al. 2011). All analyses were conducted using the Stata software (V.11.0; StataCorp, College Station, TX). p < 0.05 was considered statistically significant.

Study selection and characteristics
The systematic literature search yielded 2289 articles and 1872 were assessed for eligibility after removing duplicated papers. Among them, 1828 papers were excluded through screening titles and abstracts. For the remaining studies, 26 were excluded for the following reasons: they did not report antioxidant intake and risk of pancreatic cancer (n ¼ 10), they were review, comment or meta-analysis (n ¼ 8), lack of sufficient data (n ¼ 7) and duplicated reports (n ¼ 1). The remaining 18 studies finally met the inclusion criteria and were included in this meta-analysis (Baghurst et al. 1991;Olsen et al. 1991;Zatonski et al. 1991;Howe et al. 1992;Kalapothaki et al. 1993;Shibata et al. 1994;Ji et al. 1995;Stolzenberg-Solomon et al. 2002;Lin et al. 2005;Nkondjock et al. 2005;Stolzenberg-Solomon et al. 2009;Gong et al. 2010;Bravi et al. 2011;Amaral et al. 2012;Heinen et al. 2012;Banim et al. 2013;Han et al. 2013;Jansen et al. 2013). Figure 1 shows the flow diagram of our literature search. The characteristics of the included studies are shown in Table 1 and the methodological quality assessment is shown in Supplementary Table 1.

Association between dietary selenium intake and pancreatic cancer risk
The association between dietary selenium intake and pancreatic cancer risk was assessed by six studies. The pooled OR was 0.47 (95%CI 0.26-0.85) with significant heterogeneity (I 2 ¼82.8%, p < 0.001) ( Figure 2 and Table 2), suggesting that higher selenium intake may reduce the risk of pancreatic cancer.

Association between dietary vitamin C intake and pancreatic cancer risk
A total of 15 studies evaluated the association between dietary vitamin C intake and pancreatic cancer risk.
Among the included studies, five were prospective cohort studies and 10 were case-control studies; six studies were conducted in Europe, five were in USA and four were in other countries.
Significant heterogeneity across studies was found (I 2 ¼ 52.4%, p ¼ 0.009) ( Figure 3). Meta-regression analysis for study design, sample size and country of origin was used to explore source of heterogeneity, and we found that study design (regression coefficient ¼ 0.452, p ¼ 0.013) and sample size (regression coefficient ¼ 0.348, p ¼ 0.041) could partially explain the heterogeneity. Country of origin appeared not to be the source of heterogeneity (regression coefficient ¼ À0.092, p ¼ 0.425). We did not test the joint effect of the variables in the meta-regression analysis because of the limited number of included studies (n ¼ 15). The pooled OR for pancreatic cancer comparing the highest versus lowest categories of vitamin C intake was 0.68 (95%CI ¼ 0.57-0.80) ( Figure 3 and Table 2), indicating that higher vitamin C intake significantly decreased pancreatic cancer risk. Subgroup analyses were then applied, and we found a significant association between vitamin C intake and pancreatic cancer risk in case-control studies (pooled OR ¼ 0.58, 95%CI ¼ 0.48-0.71), while there was no significant association in cohort studies (pooled OR ¼ 0.94, 95%CI ¼ 0.78-1.15). For the subgroup analyses, results for geographic region and sample size are shown in Table 3.

Association between dietary vitamin E intake and pancreatic cancer risk
Eleven studies were included in the analysis of the association between dietary vitamin E intake and pancreatic cancer risk. Four studies were cohort studies and the remaining were case-control designed. Most of the studies were conducted in USA (n ¼ 4) or European countries (n ¼ 4), while three were in other countries. Comparing the highest versus lowest categories of vitamin E intake, the pooled OR was 0.70 (95%CI 0.62-0.81) (Figure 4 and Table 2), indicating a significant inverse association and there was no significant heterogeneity across studies (I 2 ¼ 0.0%, p ¼ 0.622) ( Figure 4).
As the subgroup analysis showed, the protective effect of vitamin E against pancreatic cancer differed by study design. Vitamin E intake was significantly associated with reduced pancreatic cancer risk for casecontrol studies (pooled OR ¼ 0.63, 95%CI 0.53-0.75) while no significant association was found for cohort studies (pooled OR ¼ 0.85, 95%CI 0.68-1.06). The subgroup analysis results for geographic region and sample size are shown in Table 3.

Association between dietary b-carotene intake and pancreatic cancer risk
The association between b-carotene intake and pancreatic cancer risk was assessed by nine studies. The pooled results suggested that higher b-carotene intake was associated with reduced pancreatic cancer risk (pooled OR ¼ 0.74, 95%CI 0.56-0.98) ( Figure 5). Significant heterogeneity among studies was found (I 2 ¼ 69.6%, p ¼ 0.001) ( Figure 5). The subgroup analysis results for study design, geographic region and sample size are shown in Table 3.

Association between b-cryptoxanthin intake and pancreatic cancer risk
Three studies were included in this analysis. Patients with highest categories of b-cryptoxanthin intake were at a lower risk of pancreatic cancer compared with the lowest categories (pooled OR ¼ 0.70, 95%CI 0.56-0.88) ( Table 2). No significant heterogeneity was observed (I 2 ¼ 28.4%, p ¼ 0.248) ( Table 2).

Association between lycopene intake and pancreatic cancer risk
Six studies were included in the pooled analysis of lycopene. The pooled OR was 0.85 (95%CI 0.73-1.00) Figure 1. Flow diagram of study selection process.   with no significant heterogeneity (I 2 ¼ 0%, p ¼ 0.701) ( Table 2), indicating a borderline significant inverse association.

Association between intake of a-carotene and pancreatic cancer risk
The association between a-carotene and pancreatic cancer risk was evaluated by four studies. We observed no significant association (pooled OR ¼ 0.86, 95%CI 0.56-1.33).

Association between lutein and zeaxanthin intake and pancreatic cancer risk
Lutein and zeaxanthin were jointly investigated by five studies and no significant association was found (pooled OR ¼ 0.82, 95%CI 0.58-1.15).

Publication bias
We did not find significant asymmetry of the funnel plots in the current study, and statistical tests for funnel plot asymmetry revealed no significant publication bias for the analysis of vitamin

Discussion
Diet has been reported to be involved in the etiology of pancreatic cancer, and vegetable and fruit consumption might be inversely associated with pancreatic cancer risk, though the evidence seemed to be limited and the protective effect was observed mainly in case-control studies (Wiseman 2008). Antioxidants were thought to be strong protective dietary components, and observational studies provided evidence for the relation between dietary antioxidants intake and pancreatic cancer risk. This meta-analysis included 18 eligible studies and evaluated the association between eight kinds of antioxidants intake and pancreatic cancer risk, including selenium, vitamin C, vitamin E, a-carotene, b-carotene, b-cryptoxanthin, lycopene, lutein and zeaxanthin. Among them, selenium, vitamin C, vitamin E, b-carotene and b-cryptoxanthin were found to be inversely associated with pancreatic cancer risk.  Subgroup analyses were then applied according to study design, geographical region, and sample size for vitamin C, vitamin E and b-carotene. Of note, inverse association was found in case-control studies while no association was found in cohort studies for these antioxidants. It is easier to conduct a case-control study than cohort study, and there were more case-control studies included in this analysis. However, case-control studies are retrospectively designed and rely on recall for antioxidant intake measurement, thus there would be more selection and recall bias. Compared with case-control studies, cohort studies could measure antioxidant intake more precisely and are better in evaluating causal relationships. Thus, more prospective cohort studies are warranted to further clarify this issue and the conclusion should be taken cautiously.
Blood levels of antioxidants can better represent the actual bioavailability and thus is an important method to assess the association between intake of micronutrients and pancreatic cancer risk. Several studies have also investigated the relation between blood antioxidant levels and risk of pancreatic cancer (Burney et al. 1989;Stolzenberg-Solomon et al. 2009;Jeurnink et al. 2014). In a small nested case-control study with 22 pancreatic cancer patients and 44 controls, pre-diagnostic serum levels of selenium and lycopene were higher among controls than cases (Burney et al. 1989). A large cohort with 29,133 male Finnish smokers examined the relation between serum a-tocopherol concentration and Table 3. Subgroup analyses of the association between vitamin C, vitamin E and b-carotene and pancreatic cancer risk.  risk of exocrine pancreatic cancer, and a significant inverse association was found (Q5 versus Q1, adjusted HR ¼ 0.52, 95%CI 0.34-0.80) (Stolzenberg-Solomon et al. 2009). Besides, Jeurnink et al. conducted a nested case-control study within the European Prospective Investigation into Cancer and Nutrition (EPIC), and found that higher plasma levels of a-tocopherol, bcarotene and zeaxanthin might be associated with reduced pancreatic cancer risk (Jeurnink et al. 2014). These studies further support the protective effects of certain antioxidants against pancreatic cancer. The mechanisms of selenium, vitamin C, vitamin E and carotenoids against pancreatic cancer are not fully understood, one explanation is their antioxidant property (McCann et al. 2005). Reactive oxygen species (ROS) are considered to be important risk factors for pancreatic cancer (Chiera et al. 2008), while antioxidants provide important defense against free-radical damage to DNA (Maritim et al. 2003). Besides, vitamin C and carotenoids were shown to have immuneenhancing role (McCullough & Giovannucci 2004). Further, experimental work indicated that the induction of apoptosis by carotenoids may be an important role of antioxidants in cancer prevention (Muller et al. 2002). Another mechanism for antioxidants is the effect against inflammatory process, while choric inflammation may play a role in carcinogenesis (Algul et al. 2007). Variations of genes regulating antioxidant pathways were reported to be associated with risk of pancreatic cancer (Wheatley-Price et al. 2008), and the inconsistence in reported association between dietary antioxidant intake and pancreatic cancer might be explained by unmeasured variation in inflammation genes or antioxidant metabolism genes (Jansen et al. 2013).
In the current study, we applied meta-regression and subgroup analyses to explore the potential source of heterogeneity. For vitamin C, meta-regression analysis suggested that study design and sample size of the included studies could partially explain the heterogeneity across studies. Other factors, including age and sex of the participants, may also be the source of heterogeneity. However, we were unable to evaluate these factors because of insufficient data. Therefore, further individual patient data (IPD) analysis would be helpful if data is available. We also observed significant heterogeneity in the analysis of selenium and b-carotene. Meta-regression analyses were not applied for the concern of limited number of studies and the heterogeneity could not be fully explained. Interestingly, we found that intake of vitamin C, vitamin E and b-carotene was not significant associated with pancreatic cancer risk in European countries, while an inverse association was observed in other geographic area (Table 3). It should be noted that most of the cohort studies reported a null association between antioxidant intake and pancreatic cancer risk, while most of them (four out of six) were conducted in European countries, which might be the reason. Besides, in the current study, Begg's funnel plot and Egger's test did not suggest significant publication bias in the pooled analysis of vitamin C and vitamin E. For the remaining meta-analyses, funnel plot showed no significant asymmetry, while no statistical tests for funnel plot asymmetry were conducted because of the possible insufficient test power. Thus, because of the limited number of the included studies, whether the publication bias exists in the current meta-analysis is still difficult to confirm.
The current analysis comprehensively assessed the association between antioxidants intake and pancreatic cancer risk, and suggested that intake of several antioxidants might be inversely associated with pancreatic risk. We applied standard meta-analysis methods and conducted the study in accordance with the PRISMA checklist, which may improve the quality of our study. Besides, most of the included studies were of high or moderate methodological quality and have relatively large sample size.
The current analysis also has some limitations. First, number of studies included in this analysis was not large enough, thus some of the subgroup analyses were hard to conduct and less reliable. Second, most of the involved studies were case-control studies which were prone to bias. Besides, the included studies were mainly from Europe, USA and Asia, therefore the conclusions should be taken cautiously for other ethnic populations.

Conclusions
In summary, higher dietary intake of antioxidants including selenium, vitamin C, vitamin E, b-carotene and b-cryptoxanthin, was inversely associated with pancreatic cancer risk. More studies, especially prospective cohort studies are warranted to further clarify this issue.