Vitamin Intake Reduce the Risk of Gastric Cancer: Meta-Analysis and Systematic Review of Randomized and Observational Studies

Aim The association between vitamin intake and gastric cancer (GC) has been widely debated due to the relatively weak evidence. In this study, a meta-analysis of prospective and well designed observational studies were performed to explore this association. Methods MEDLINE, Cochrane Library, and Sciencedirect were searched for studies of vitamin consumption and gastric cancer. This produced 47 relevant studies covering 1,221,392 human subjects. Random effects models were used to estimate summary relative risk (RR). Dose-response, subgroup, sensitivity, meta-regression, and publication bias analyses were conducted. Results The RR of gastric cancer in the group with the highest vitamin intake was compared to that of the lowest intake group. Total vitamin intake was 0.78 (95% CI, 0.71−0.83). In 9 studies that individuals were given doses at least 4 times above the tolerable upper intake (UL) vitamins, the RR was 1.20 (95% CI, 0.99−1.44). However, in 17 studies that individuals received doses below the UL, the RR was 0.76 (95% CI, 0.68−0.86). Dose-response analysis was conducted on different increments in different types of vitamins (vitamin A: 1.5 mg/day, vitamin C: 100 mg/day, vitamin E: 10 mg/day) intake with a significant reduction in the risk of gastric cancer, respectively, 29% in vitamin A, 26% in vitamin C, and 24% in vitamin E. Conclusion This meta-analysis clearly demonstrated that low doses of vitamins can significantly reduce the risk of GC, especially vitamin A, vitamin C, vitamin E.


Introduction
Gastric cancer (GC) is the second leading cause of cancer-related mortality worldwide, with an estimated 989,600 new cases and accounted for 738,000 deaths in 2011. [1]. Despite the decrease in overall incidence, the total survival rate for GC patients did not improve significantly over the past two decades [2].The only potentially curative treatment for GC is surgery, but only about 20-40% of patients can undergo radical resection. GC have become the main contributors to the total cancer burden in many parts of Asia [3]. Effective primary prevention strategies for GC, especially vitamin intake, have drawn considerable attention. For example, vitamins have been reported to play an important role in the prevention of GC in many studies [4,5]. Some in vitro studies have also suggested that vitamins may prevent GC through different processes, such as scavenging the concentration of nitrite in the stomach, reducing oxidative stress, and inhibiting nitrosation.
Since 1970 s, the association between vitamin intake and GC has been assessed in a large and rapidly expanding body of literature. [6][7][8] However, most RCTs (Randomized, Placebo-Controlled Trials) included were not designed primarily to investigate the relationship between vitamins consumption and GC and performed in high-risk individuals. The current study is the first high-quality analysis of both prospective and retrospective studies to explore the relationship between vitamin intake and the riskof GC.

Search Strategy and Study Selection
MEDLINE, Cochrane Library and Sciencedirect were searched for studies of vitamin consumption and GC that were published only in English and performed on human participants from inception to February 2, 2014. Search terms were as follows: (vitamin OR supplement OR food OR diet OR dietary) AND (gastric OR stomach) AND (cancer OR neoplasm OR carcinoma). The reference lists of the articles identified were scanned manually for further potentially relevant studies. Authors were asked if they knew of any useful additional information (S1 Table  and S2 Table in S1 File).
A study was included if it met the following criteria: 1) original article; 2) placebo-control, case-control or cohort design; 3) vitamin intake as the exposure of interest; 4) GC occurrence provided; 5) odds ratio (OR) or RR, and the corresponding 95% confidence interval (CI). Animal, mechanistic studies and non-peer-reviewed articles were excluded. This meta-analysis was performed in accordance with the Preferred Reporting Items for Meta-Analyses (PRISMA) statement checklist (checklist in checklist S1).

Data Extraction and Quality Assessment
Four authors independently assessed the retrieved studies and extracted all data according to the pre-specified selection criteria. Disagreements were resolved by discussion. The following information was collected from each study: the last name of the first author, year of publication, study design, location, participant age, participant sex, study period, type of control subjects in case-control studies, sample size, type of vitamins evaluated and type of intake, the OR or RR with corresponding 95%CI for each category, and adjustments for confounders. When several articles discussed the same study, only the most recent or the one with the most complete data was included here. An evaluation system based on the Newcastle-Ottawa scale (NOS) was used to estimate the quality of observational studies. The studies included here were evaluated for three major factors: selection, comparability, and exposure/outcome assessment. The perfect score was 10 stars, and studies with 7 or more stars were defined as high-quality. Due to the risk of overestimation of beneficial intervention effects RCTs of low or inadequate methodological quality, we also assessed the RCTs methodological quality from the following domains: allocation sequence, allocation concealment, blinding, follow-up, and other apparent biases.

Statistical Analysis
All analyses were performed with Rev Man version 5.2 and STATA 12.0. P,0.05 was defined as significant. ORs or RRs were extracted from the studies included here, and their standard errors (SEs) were calculated from their respective CIs. A random-effects model was used to quantify the relationship between vitamin intake and the risk of GC, considering both intra-and inter-study variability (t 2 ). The measure of effect of interest was RR with 95% CI. Because the absolute incidence of GC was low, the RR was mathematically similar to the OR in the studies included here. For this reason, all results were reported as RR for simplicity. Heterogeneity among studies was evaluated withx 2 and I 2 statistical testing. [9] To assess heterogeneity across all included studies, the variables of study design, geographic area, method of evaluation of vitamin intake, and dose were further examined in a meta-regression model. Subgroup stratification analyses were performed to assess variations in influence of these variables on overall results. Because the characteristics of the subjects, method of assessment of vitamin intake, and adjustments for confounders differed across studies, a sensitivity analysis was performed to assess any possible causes of heterogeneity and to evaluate the impact of different exclusion criteria on overall outcome. The influence of each single study on the results was evaluated by removing each study from consideration one at a time.
For the dose-response meta-analysis, only studies that listed the following data were analyzed: number of the case and control subjects, examined RR or OR and their 95% CI, and at least three quantitative exposure categories. For each included study, the mean vitamin intake for each quantitative exposure category was assigned an RR. Publication bias was assessed using funnel plots and Egger's test method [10,11].
Quality scores of observational study are summarized in S4 Table and S5 Table in S1 File. Quality scores ranged from 7 to 10. The average score was 8 for casecontrol studies and cohort studies. In this way, all observational studies were found to be high quality according to the NOS evaluation system. RCTs quality scores were also evaluated in S6 Table in S1 File. Twenty-two studies were excluded because they did not report usable data. Four papers were excluded because they reported the same study. Eight studies were excluded because they did not investigate the association between vitamin intake and GC risk. Noncohort studies and 142 reviews were also excluded.

3). Vitamin Dose
Analysis by vitamin dose showed dosage (low dose) less than UL to be associated with lower risk of GC (Fig. 4). In 9 studies (n 5152,848), individuals were given doses at least 4 times above the UL (high dose), and       Table 2).

Sensitivity Analyses and Meta-regression
Sensitivity analyses were conducted to explore possible causes of heterogeneity and the effect of various exclusion criteria on the overall result were examined (data not shown). Sixteen studies that were not adjusted for total energy intake or dietary factors were omitted. [ Meta-regression analysis demonstrated that study design (P 50.075), vitamin dosage (P 50.006), and method of assessing vitamin intake (P 50.006) were significant sources of heterogeneity. Study design alone explained 8.49% of the t 2 in the meta-regression analyses, vitamin dosage explained 24.54% of the t 2 and assessment of vitamins intake explained 23.84% (S8 Table in S1 File).

Publication Bias
The funnel plot did not show any obvious asymmetry (S6 Figure in S2 File). No publication bias was detected using the Egger's test (P 50.254).

Discussion
In this study, data were available for more than 1.2 million individuals and more than 11,000 GC events. This work provided convincing evidence that vitamins intake is associated with a reduced risk of GC, especially at low doses. This relationship between vitamin intake and GC risk was apparent and consistent across a wide range of stratified subgroups. The dose-response meta-analysis indicated that appropriate increase vitamins intake (vitamin A: 1.5 mg/day, vitamin C: 100 mg/day, vitamin E: 10 mg/day) were associated with a statistically significant decreased risk of GC: 36% in vitamin A, 35% in vitamin C, and 32% in vitamin E, respectively.
In fact, since 1970 s, many observational studies and RCTs have evaluated the relationship between vitamin intake and the risk of GC, though results have been mixed. Zheng and Carman have provided evidence that higher vitamin intake may be relevant to the prevention of cancers of the upper digestive organs. [59,64] A interesting study from China also reported higher circulating vitamin was associated with a reduced risk of incident GC [65]. However, Other investigators concluded that supplementation with vitamins has no major impact on the occurrence of GC [49,55]. The discrepancy has several possible explanations, including differences in study design and type of vitamin intake (dietary or supplemental), differences in vitamin dosage used, differences in the assessment of vitamins intake and potential biases in each study. The lack of a statistically significant outcome in the clinical trials may have been caused by any of several methodological limitations of trials, such as short follow-up period and high levels of vitamins used.
Several meta-analyses of RCTs have also analyzed the effect of vitamins on the prevention of gastrointestinal cancer [66][67][68][69]. Wu revealed that vitamin A intake was inversely associated with GC risk by a meta-analysis, [66] while other researchers came to a opposite conclusion. They found that antioxidant vitamins supplements cannot prevent GC, and may even increase overall mortality [67][68][69]. However, there were many limitations in these meta-analyses. Firstly, the RCTs included in previous meta-analyses had higher doses than those usually found in individuals who ate a balanced diet, and some trials used dosages well above the recommended UL. [7, 40, 44, 45, 48, 50-52, 55, 56] (S9 Table in S1 File) The doses used in this study are more reasonable. Secondly, in prior articles, many retrospective case-control studies on this topic were excluded, despite which showed strongly that vitamins intake can prevent GC. In fact, most RCTs included in previous meta-analyses were not designed primarily to investigate the relationship between vitamins consumption and GC. This led to a lack of adjustment for the main confounders of GC. Moreover, most of these RCTs were performed in high-risk individuals, such as longtime smokers, [40,44,50,55,56] and subjects with a history of premalignant lesions, [8,42,54] which may not reflect the vitamin intake of normal risk population. Thus, the total number of subjects of previous meta-analyses was not very substantial and their conclusions should be treated with caution. This paper includes discussion of many well designed observational studies. These were conducted in normal risk populations, and are closely related to the topic. Indeed, it should not be assumed that RCTs always provide high-quality evidence for therapy. [70] High-quality observational studies are also important sources of powerful evidence in meta-analyses. [71].
Some studies have reported other non-antioxidant vitamins' that affect GC prevention, [8,33,39,54] others have focused on antioxidant vitamins (vitamin A, vitamin C and vitamin E). [45,53,56] However, in daily diet, it is difficult to draw distinctions between non-antioxidant vitamins and antioxidant ones. In this study, we combine them and demonstrate vitamins intake can reduce risk of gastric cancer.
The results of this meta-analysis indicate that relatively low doses of vitamins can prevent the occurrence of GC. Dose and method of administration are often clinically important and can be manipulated to prevent cancer [72]. For example, in the famous ATBC clinical trial, [56] the long-term use of vitamin A (4 years) at a high dose (7.5 mg/day, about 2.5 times the UL) showed no benefit with respect to preventing lung cancer in high-risk individuals (smokers). However, in a HCC study conducted in southwestern France, the author emphasized that dietary vitamin A (2 mg/day, less than the UL) might have a distinct and important protective effect on lung cancer prevention. [73] Some high-quality retrospective analyses indirectly showed that relatively low doses of vitamins (less than UL) prevented cancer more effectively. [74] These conclusions are similar to our study. Notably, in the dose-response analysis, we revealed that relatively low doses vitamin A, vitamin C, and vitamin E can significantly reduce the risk of GC (vitamin A: 1.5 mg/day, vitamin C: 100 mg/day, vitamin E: 10 mg/day). They are hopeful to be a possible recommendation dosage of vitamin intake for GC prevention. However, the mechanism of low doses of vitamins reduce risk of cancer is still unknown. Some researchers have also revealed that the long term administration of mega-dosages of vitamins can bring out many adverse effects.
The current study also draws attention to the fact that vitamins from food (plant or animal) contribute more to reductions in GC risk than synthetic vitamin supplements. Some investigators have noted that the bioavailability of vitamins differs depending on whether the vitamin comes from food or is synthetic, which could explain the results. For example, Carr reported differences in bioavailability between synthetic and kiwifruit-derived vitamin C in a randomized crossover pharmacokinetic study [75].
Subgroup analyses by vitamin types, vitamin A, vitamin B, vitamin C and vitamin E produced similar outcomes, but vitamin D did not. Vitamin D is not really a vitamin. It is the precursor to the steroid hormone calcitriol and play an important role in determining cancer risk [76]. Accumulating results from preclinical and clinical studies strongly suggest that vitamin D deficiency increases the risk of developing cancer. Vitamin D supplements might be an economical and safe way to reduce the incidence of cancer and improve cancer prognosis and outcome. However, in the current meta-analysis, only 5 case-control studies have explored the association between vitamin D and GC risk [8,20,29,37,39]. This might be the reason for the discrepancy.
During the past 3 decades, many studies have reported that the mechanisms of different types of vitamins may reduce the risk of GC. This includes vitamin that function in an irreversibly oxidized form, vitamins that reduce the concentration of nitrite in the stomach, and vitamins that affect free radical-mediated damage to the stomach epithelium [75]. In addition, some studies have indicated that vitamin E is a potent lipid-soluble antioxidant and might be involved in GC prevention by reducing oxidative stress [77].

Study Strengths and Limitations
The current study has several strengths. First, it addresses both non-antioxidant and antioxidant vitamins and covers a large number of human subjects (1,221,392). This increased the statistical power of the analysis considerably. Second, these results are less likely to be explained by recall and selection bias because of the inclusion of 18 prospective studies (11 RCTs and 7 cohort studies). Third, a statistically significant association was observed in most of the subgroups that adjusted for confounders. These subgroups produced results similar to those of other subgroups. Fourth, the current study not only included RCTs but also many other high-quality observation studies. This was beneficial to identify the relationship between vitamins and GC. Fifth, a significant dose-response relationship was observed between vitamin intake and GC risk (Table 2). Finally, this is the first study to discuss the influence of dosage in the relationship and the effect of all kinds of vitamin compare with early studies.
Several limitations should be addressed in this study. First, the studies included in this article have been conducted in different countries since the 1980 s, but some studies have had faulty designs, were not designed primarily to study vitamins consumption, and lacked stratification. This makes the combination of these studies with a random-effects model problematic. The second limitation is that the quality and power of any meta-analysis are dependent on the quality and comparability of data from the included studies. The analysis would be more convincing if original data were available, making an adjustment estimate possible. We have attempted to contact the authors of original studies to obtain more detailed information. However, it is very difficult to obtain all the original data regarding published studies. Third, the range of vitamin taken in by individuals with the lowest vitamin intake and those with the highest differed among the studies, which caused heterogeneity in the pooled analysis. Fourth, there were relatively few eligible studies of the dose-response analysis. These studies contained a few cohort and case-control studies. More and more in-depth studies are necessary.

Implications
The current findings may have several implications. First, vitamin intake can reduce the risk of GC, but excessive and long-term intake might disturb this antitumor function. Second, dietary vitamins might prevent GC more effectively than supplements. Third, according to the results of the current meta-analysis, overall vitamin intake can reduce the risk of GC by 23%. This reduction could be translated into a decrease of as many as 169,740 GC deaths and 227,608 new cases per year worldwide [1]. Last, the desired low but sufficient level of vitamin intake may be achieved by fruit and vegetable consumption. This is consistent with results indicating fruit and vegetable intake is inversely associated with the incidence of GC [78].

Conclusions
In summary, unlike early studies, this article conducted well designed observational studies which conducted in normal risk populations and discuss the influence of dosage in the relationship and the effect of all kinds of vitamins. It shows clearly that low doses of vitamins can significantly reduce the risk of GC, especially vitamin A, vitamin C, vitamin E. However, because of potential bias and confounding factors, these results should be treated with caution. More and better-designed large clinical trials should use appropriate doses of vitamins in order to generate a more visible association between vitamin intake and the risk of GC.
Supporting Information S1 PRISMA Checklist. Preferred Reporting Items for Meta-Analyses (PRISMA) statement checklist. doi:10.1371/journal.pone.0116060.s001 (DOC) S1 File. Supporting Information Tables. S1 Table Search strategy in PubMed and Cochrane Library. S2 Table. Search strategy in Sciencedirect. S3 Table. Characteristics of the included studies. S4 Table. Methodological quality of casecontrol studies included in the meta-analysis. S5 Table. Methodological quality of cohort studies included in the meta-analysis. S6 Table. Methodological quality of RCTs included in the meta-analysis. S7 Table. Dose-response analysis. S8 Table. Meta-regression analysis. S9 Table. Tolerable upper intake levels of vitamins. doi:10.1371/journal.pone.0116060.s002 (DOCX) S2 File. Supporting Information Figures. S1 Figure. Subgroup analysis: Forest plot of vitamin type. CI, confidence interval; df, degrees of freedom; I2, the percentage of total variation across studies that is caused by heterogeneity rather than by chance Squares or diamonds to the left of the solid vertical line indicate benefit with each type of vitamin intake; this is conventionally significant (P,0.05) only if the horizontal line or diamond does not overlap the solid vertical line. Relative risks are analysed with random-effects model. S2 Figure. Subgroup analysis: Forest plot of Lauren's classification (intestinal). CI, confidence interval; df, degrees of freedom; I 2 , the percentage of total variation across studies that is caused by heterogeneity rather than by chance. S3 Figure. Subgroup analysis: Forest plot of Lauren's classification (diffuse). CI, confidence interval; df, degrees of freedom; I 2 , the percentage of total variation across studies that is caused by heterogeneity rather than by chance. S4 Figure. Subgroup analysis: Forest plot of location (cardia). CI, confidence interval; df, degrees of freedom; I 2 , the percentage of total variation across studies that is caused by heterogeneity rather than by chance. Relative risks are analysed with random-effects model. S5 Figure. Subgroup analysis: Forest plot of location (non-cardia). CI, confidence interval; df, degrees of freedom; I 2 , the percentage of total variation across studies that is caused by heterogeneity rather than by chance. Relative risks are analysed with random-effects model. S6 Figure.