Association between vitamin A, retinol and carotenoid intake and pancreatic cancer risk: Evidence from epidemiologic studies

Pancreatic cancer is a devastating disease with poor prognosis. The association between vitamin A, retinol and carotenoid intake and the risk of pancreatic cancer occurrence remains controversial, and therefore it is necessary to make a meta-analysis to clarify the association between vitamin A, retinol and carotenoid intake and pancreatic cancer risk. In the present study, PubMed and EMBASE databases were used to identify qualified studies. The association between dietary vitamin A, retinol and carotenoids was estimated by pooled odds ratios (ORs) and corresponding 95% confidence intervals (CIs). It was found that there was an inverse correlation between vitamin A, beta-carotene and lycopene intake and the risk of pancreatic cancer (for vitamin A, pooled OR = 0.85, 95%CI = 0.74–0.97, P = 0.015; for beta-carotene, pooled OR = 0.78, 95%CI = 0.66–0.92, P = 0.003; for lycopene, pooled OR = 0.84, 95%CI = 0.73–0.97, P = 0.020), which was more prominent in case-control study subgroup. In conclusion, dietary vitamin A, beta-carotene and lycopene might inversely correlate with pancreatic cancer.

Pancreatic cancer is a devastating disease with poor prognosis and the 5-year survival rate remains low at 8% 1 . It is the eighth and ninth leading cause of cancer-related death in men and women respectively throughout the world 2 . For patients with resectable pancreatic cancer, surgery is the mainstay of treatment. But the median overall survival time remains low in all pancreatic cancer stages 3 . There have been few therapeutic advances or effective treatments over the last few years 4 , highlighting the importance of identifying preventive factors for this malignancy. Risk factors such as smoking, obesity, diabetes mellitus, chronic pancreatitis and established genetic syndromes are known to be associated with pancreatic cancer 5 . A number of epidemiologic studies have been published in an attempt to explore the relationship between nutrient intake and the risk of pancreatic cancer occurrence. Various vitamins including vitamin B 6 , vitamin C 7 and vitamin E 8 have been implicated in the risk of pancreatic cancer occurrence according to previous studies.
Vitamin A (retinol) and its derivatives are a group of fat soluble compounds composed of a similar structure which are rich in cod liver oil and play important role in multiple biological processes 9 . Due to their ability to promote normal embryonic development and exert effects on cellular differentiation, they are essential for all stages of life from embryogenesis to adulthood 10 . However, they cannot be synthesized de novo by animals (including human) and must be obtained from the diets 11 . Recently, a myriad of epidemiological studies have demonstrated an inverse relationship between dietary vitamin A consumption and cancer development 12 . For instance, vitamin A has been proved to play a protective role in breast cancer 13 and lung cancer 14 . However, the association between vitamin A (including retinol and carotenoid) and pancreatic cancer remains controversial [15][16][17][18] . Zablotska et al. conducted a case-control study to evaluate the association of dietary vitamin D, calcium and retinol and the risk of pancreatic cancer in USA, finding that there was no signification association between them 18 . Also, Kalapothaki et al. found that vitamin A intake was not related to pancreatic cancer risk when crude fiber intake was adjusted 16 . The results of clinical studies are not consistent with those of molecular researches. But other carotenoids, such as lycopene, alpha-and beta-carotene, are associated with pancreatic cancer risk 17,19 . Therefore, a meta-analysis is necessary to clarify the association between vitamin A, retinol and carotenoids intake and pancreatic cancer risk.
Quantitative synthesis. Vitamin A and pancreatic cancer. The results pooled by the fixed effect model indicated that there was an inverse association between vitamin A intake and pancreatic cancer risk (OR = 0.85, 95%CI = 0.74-0.97, P = 0.015) (Fig. 2). In addition, stratification analysis conducted by ethnicity and study design type revealed a significant association between vitamin A intake and pancreatic cancer risk in Caucasians subgroup (OR = 0.84, 95%CI = 0.73-0.96, P = 0.011) and case-control subgroup (OR = 0.83, 95%CI = 0.72-0.95, P = 0.007). Subsequently, publication bias was test by funnel plot and Egger's test. The Egger's test value suggested that significant publication bias was in the meta-analysis (P = 0.052). The results of metatrim suggested that the summary OR was 0.815 and corresponding 95%CI was 0.702 to 0.946. Besides, no single study could change the results in sensitive analyses, implying that the results of this meta-analysis were robust.
Retinol and pancreatic cancer. The meta-analysis based on 11 studies of 9 articles indicated that there was no significant correlation between retinol intake and pancreatic cancer risk (OR = 1.02, 95%CI = 0.78-1.34, P = 0.860). Subgroup analysis by ethnicity and the results showed no significant correlation between retinol intake and the risk of pancreatic cancer (Fig. 3). Additionally, the stability of the results was estimated by sensitive analysis, showing that a good stability and credibility. Publication bias was also tested by funnel plot and Egger's test (P = 0.591), suggesting that there was no statistically significant publication bias in this meta-analysis.
Lycopene and pancreatic cancer. As shown in Fig. 8, there was an inverse correlation between lycopene intake and pancreatic cancer risk (OR = 0.84, 95%CI = 0.73-0.97, P = 0.020). When stratified by ethnicity, there was an inverse relationship between lycopene intake and pancreatic cancer risk in Caucasians (OR = 0.86, 95%CI = 0.73-1.00, P = 0.05), while this correlation was insignificant in the mixed population (OR = 0.78, 95%CI = 0.54-1.13, P = 0.187). With respect to the study design type, decreased the pancreatic cancer risk in case-control study (OR = 0.77, 95%CI = 0.64-0.92, P = 0.005), while prospective study showed no association between lycopene intake and pancreatic cancer (OR = 0.98, 95%CI = 0.78-1.23, P = 0.844). The sensitive analysis revealed that the results of this meta-analysis were credible and stable. Publication bias was evaluated by forest plot and Egger's test (P = 0.857), and the results showed no significant publication bias in the meta-analysis.
Cryptoxanthin and pancreatic cancer. The results showed no significant association between cryptoxanthin intake and pancreatic cancer risk (OR = 0.86, 95%CI = 0.67-1.12, P = 0.276). After stratification by ethnicity, there was no significant association in Caucasians and the mixed population. Similarly, there was no significant correlation in Caucasians and the mixed population. Subgroup analysis showed no significant correlation between cryptoxanthin intake and pancreatic cancer risk in case-control study and prospective study. The sensitive analysis implied that the result of the meta-analysis was robustness. The forest plot and Egger's test (P = 0.522) suggested that no significant publication bias in this meta-analysis.
Lutein and zeaxanthin and pancreatic cancer. Pooled ORs and corresponding 95%CIs indicated that there was no significant correlation between lutein and zeaxanthin intake and pancreatic cancer risk (OR = 0.80, 95%CI = 0.61-1.05, P = 0.104). As it showed in Fig. 9, there was an inverse association between lutein and zeaxanthin intake and pancreatic cancer risk in the mixed population (OR = 0.61, 95%CI = 0.42-0.89, P = 0.010), but not in Caucasians population (OR = 0.84, 95%CI = 0.62-1.3, P = 0.251). Besides, subgroup analysis showed no significant association could be found in the subgroup of case-control study and prospective study. The sensitive analysis implied that this result was robust. Furthermore, the forest plot and Egger's test (P = 0.664) showed no significant publication bias in our meta-analysis. In summary, there was an inverse correlation between vitamin A (including some carotenoids) intake and pancreatic cancer risk, but no significant correlation was observed between retinol intake and pancreatic cancer risk. All the results were summarized in Table 2.

Discussion
This meta-analysis included 18 articles focusing on the correlation between vitamin A, retinol and carotenoid intake and pancreatic cancer risk. The result showed that dietary vitamin A, carotene, beta-carotene and lycopene were inversely correlated with the risk of pancreatic cancer risk. However, retinol, alpha-carotene, cryptoxanthin, lutein and zeaxanthin intake had no relationship with pancreatic cancer risk. Vitamin A is a necessity for cell growth and differentiation of epithelial tissues and must be obtained from diets in the human body. Provitamin A compounds, such as beta-carotene can transform into vitamin A, which is an essential molecule entailing multiple developmental pathways and influencing cell proliferation and differentiation in a variety of cell types 21 . Molecular studies had demonstrated that retinoids (vitamin A and its metabolites) could cause apoptosis in pancreatic cancer cells and thus suppress pancreatic cancer growth via activation of retinoic acid receptor-gamma, suggesting that vitamin A and its metabolites may play a protective role against pancreatic cancer 22 . Additionally, several preclinical studies showed that retinols play roles in many signaling pathways related with cell growth, adhesion and migration [23][24][25] . A recent study revealed that retinoic acid could inhibit pancreatic cancer cell migration and epithelial-mesenchymal transition by decreasing the expression of  interleukin 6 (IL-6) in cancer-associated fibroblast (CAFs) cells 25 , suggesting that retinoids could be applied for prevention or therapy the recurrence and metastasis of pancreatic cancer. Actually, immunotherapy including 13-cis-retinoic acid and interleukin 2 had been used for treating locally advanced pancreatic cancer 26 . However, there is no clinical study focusing on vitamin A therapy in pancreatic cancer so far.
Other carotenoids such as lycopene and zeaxanthin that cannot convert into vitamin A may act as antioxidants against cancer initiation and progression. Lycopene has been proved to be a potent inhibitor for cell proliferation  and growth in some cancer cells, such as endometrial cancer, breast cancer and lung cancer 27,28 . More recently, Assar and colleagues reported that lycopene could suppress the nuclear factor kappa B (NF-κ B) signaling pathway through inhibiting phosphorylation of inhibitor of kappa B (Iκ B) in human prostate and breast cancer cells, probably due to the action of lycopene as an antioxidant to scavenge free radicals 29 . These data provide a potential strategy to prevent and treat pancreatic cancer by using lycopene.
Several meta-analyses or pooled analyses have investigated the association between the intake of other vitamins and pancreatic cancer risk. Fan et al. conducted a meta-analysis to assess the relationship between dietary vitamin C and pancreatic cancer risk and found that a higher vitamin C intake was inversely correlated with pancreatic cancer risk 7 . Similarly, dietary vitamin E was found to be a protective factor against pancreatic cancer 8 . However, a recent pooled analysis suggested that higher levels of vitamin D might increase pancreatic cancer risk 30 . To the best of our knowledge, this is the first meta-analysis about the relationship of dietary vitamin A, retinol, and carotenoids with pancreatic cancer risk. Overall, we found that dietary vitamin A had an inversely association with pancreatic cancer risk. On the contrary, several studies had investigated the relationship between vitamin A intake and the risk of pancreatic cancer and the results were negative. Partly, these conclusions may due to the small sample size of each study. Besides, the only prospective study showed no association between vitamin A intake and the risk of pancreatic cancer 31 . However, this prospective study was based on male smokers instead of the general population. No significant relationship was found between dietary retinol and pancreatic cancer risk, or in the subgroups of Caucasians, Asians and the mixed population. These results are consistent with previous case-control studies 15,17,[32][33][34][35] .
The association between carotenoid intake and cancer risk had been investigated in many cancer types. Zhou et al. conducted a meta-analysis and found that beta-carotene and alpha-carotene were inversely correlated with risk of gastric cancer 36 . Another meta-analysis revealed that dietary alpha-carotene and lycopene could decrease the risk of prostate cancer 37 . In our meta-analysis, beta-carotene and lycopene intake were inversely associated  with pancreatic cancer risk, while alpha-carotene and cryptoxanthin intake had no significant relationship with pancreatic cancer risk. Many previous observational studies including 10 case-control studies 15,17,19,32,33,[38][39][40][41] and 4 prospective studies 31,42-44 on the relationship between beta-carotene intake and pancreatic cancer risk reported inconsistent results. These discrepant results from case-control studies might result from recall bias of self-reported dietary intake and different ethnicity. However, the results of 4 prospective studies did not suggest that beta-carotene acted as a protective factor against pancreatic cancer. Notably, a nested case-control study performed in Europe indicated that higher plasma concentrations of beta-carotene could decrease the risk of suffering from pancreatic cancer 45 . Besides, this article also suggested that higher plasma concentrations of zeaxanthin might be inversely related to pancreatic cancer risk, which is consistent with the result of our meta-analysis indicated that the relationship between the total lutein and zeaxanthin intake and pancreatic cancer risk in the subgroup of the mixed population. However, this subgroup result was based on only one case-control study 38 .
Between-study heterogeneity is common in meta-analysis 46 , and the heterogeneity test showed moderate between-study heterogeneity in most meta-analyses. The study characteristics of each study may lead to heterogeneity. In our meta-analysis, the study design, geographic location, publication year and sources of control and cases are various. To find the causes of heterogeneity for covariates, we conducted a meta-regression and   subgroup analysis. As a result, meta-regression failed to determine any study characteristics including publication year, study type, study size and ethnicity as sources of heterogeneity. Subgroup analyses by study type and ethnicity were performed to explore the source of heterogeneity. However, between-study heterogeneity was permanent in some subgroups, suggesting that other unknown confounding factors may be present. In addition, the adjustments and the intake levels of nutrients are different between these studies. Although the results obtained in our meta-analysis are statistically significant as a whole, several limitations should be noted in interpreting our study. First, most included studies in our meta-analysis were case-control studies, in which recall bias may be unavoidable. Both case-control and cohort study are observational study, they require fewer resources but provide less evidence compared with randomized controlled trial (RCT). However, given the extremely low morbidity of pancreatic cancer, there is too difficult to conduct RCT on the association between vitamin A, retinol and carotenoid intake and the risk of pancreatic cancer. Second, some confounding factors such as eating habits and residual confounding cannot be measured, which may affect the stability and credibility of our meta-analysis. Third, the individual sample sizes for each case in most studies included in this meta-analysis were relatively small and these studies were conducted in different populations whose heredity might be different. Finally, some pancreatic cancer cases may be familial heredity 47 , which may change the morbidity of pancreatic cancer and lead to an inaccurate results in epidemiological study.
In conclusion, the results of our meta-analysis indicate that high-level vitamin A, carotene, beta-carotene and lycopene intake might be the potential factors related to low pancreatic cancer risk. However, due to the limitations of the present meta-analysis mentioned above, it should be prudent to make recommendations based on the results of the present meta-analysis.

Materials and Methods
Literature search strategy. PubMed and EMBASE databases were used to identify observational studies that reported the association between vitamin A, retinol and carotenoid intake and the risk of PANCREATIC CANCER up to December 30 th , 2015 by using the following key words "Vitamin A or Vitamin or diet or dietary or retinol or carotenoids or carotene or cryptoxanthin or lycopene or lutein or zeaxanthin" and "pancreatic" and "cancer or carcinoma or neoplasm or tumor or adenocarcinoma". Additionally, some potential studies were identified via secondary searches which were conducted by searching reference lists of selected literatures.
Inclusion and exclusion criteria. The inclusion criteria were: (1) observational studies including case-control and cohort study design; (2) studies reporting the association between exposure factors including vitamin A, retinol and carotenoids and the risk of pancreatic cancer; (3) studies published in English or Chinese; (4) Providing the odds ratio (OR) (or relative risk [RR], hazard risk [HR]) data and the corresponding 95% corresponding interval (CI) for the highest vs. the lowest level of vitamin A intake or retinol intake or other carotenoid intake. The exclusion criteria were: (1) reviews, meta-analyses, case reports, editorials or human-uncorrelated experiments; (2) duplicated study (If duplicated studies were present, the study with the largest sample size was selected); (3) studies not reporting OR(or RR, HR) and 95%CI or lacking sufficient data to calculate OR(or RR and HR) and 95%CI. Data extraction. The process of data extraction was conducted by two authors independently with a standardized form based on the inclusion and exclusion criteria mentioned above. Any divergence was resolved by rechecking until consensus was reached. The following information was collected: the last name of the first author, publication year, country, ethnicity, study design, number of cases and controls or total sample size, carotenoid types, OR(or RR, HR), the corresponding 95% CI from the most fully adjusted model for the highest vs. the lowest vitamin A intake, and the factors of adjustment for covariates. Quality assessment. Newcastle-Ottawa-Scale (NOS) was applied in quality assessment 48 . During this process, the quality of the selected observational studies was evaluated independently by two authors. The NOS is a nine-point scale containing three parts: selection (four points), comparability (two points) and exposure/outcome assessment (three points). A study with a NOS score ≥ 6 was regarded as a high-quality study, and vice versa. Statistical analysis. If the outcome under study is rare in all populations and subgroups under review, one can generally ignore the distinctions between the various measures of relative risk 49 . Given the low absolute risk of pancreatic cancer in the general populations, we interpreted all risk estimates as OR for simplicity. The relationship between vitamin A, retinol and other carotenoids and the risk of pancreatic cancer was assessed by calculating pooled OR and 95%CI respectively. Additionally, subgroup analysis was conducted by study design type and ethnicity if sufficient data was provided. All the statistical tests were two-sided and the results were considered as statistically significant if P ≤ 0.05. The heterogeneity test was performed by Q test and I 2 . When I 2 > 50%, the random effect model was suggested to calculate the pooled OR and 95%CI. Otherwise, the  Table 2. The results of the association between vitamin A, retinol and carotenoids intake and the risk of pancreatic cancer in meta-analysis. P a : P value for meta-analysis; P b : P value for heterogeneity test.
Scientific RepoRts | 6:38936 | DOI: 10.1038/srep38936 fixed effect model was applied. Furthermore, sensitivity analysis was conducted by omitting each study once a time. We also assessed publication bias via funnel plots and Egger's test 50 . This meta-analysis was performed by STATA12.0 (STATA Corporation, College Station, TX).