Association of Risk of Gastric Cancer and Consumption of Tobacco , Alcohol and Tea in the Chinese Population

Gastric cancer (GC) has become one of the most serious diseases threatening human health and life all over the world. GLOBOCAN 2008 reported that there were about 989 thousand new GC cases (including 640 thousand males and 349 thousand females) and 738 thousand GC deaths (including 464 thousand males and 273 thousand females) worldwide (Ferlay et al., 2010). The disease is also prevalent in China. It ranked the third most common cancer in Chinese population and estimated new cases of and deaths from GC were about 400 and 300 thousand respectively per year (Yang et al., 2005; Yang, 2006). Although it can be said that GC (like most other types of cancer) is a multi-factor-induced disease and is attributable to both genetic and environmental factors, our understanding of the causes of GC is generally limited (Guggenheim et al., 2013). H. pylor infection is perhaps the most widely recognized cause of GC (Uemura et al., 2001; Helicobacter and Cancer Collaborative Group, 2001; Wroblewski et al., 2010). Yet it accounts for


Introduction
Gastric cancer (GC) has become one of the most serious diseases threatening human health and life all over the world.GLOBOCAN 2008 reported that there were about 989 thousand new GC cases (including 640 thousand males and 349 thousand females) and 738 thousand GC deaths (including 464 thousand males and 273 thousand females) worldwide (Ferlay et al., 2010).The disease is also prevalent in China.It ranked the third most common cancer in Chinese population and estimated new cases of and deaths from GC were about 400 and 300 thousand respectively per year (Yang et al., 2005;Yang, 2006).Although it can be said that GC (like most other types of cancer) is a multi-factor-induced disease and is attributable to both genetic and environmental factors, our understanding of the causes of GC is generally limited (Guggenheim et al., 2013).H. pylor infection is perhaps the most widely recognized cause of GC (Uemura et al., risk factor for numerous cancers worldwide, including cancer of the oral cavity and pharynx, esophagus, stomach, larynx, colorectum, central nervous system, pancreas, breast and prostate (Boffetta et al., 2006;de Menezes et al., 2013).Others have demonstrated that the first metabolite (acetaldehyde) of alcohol is a local carcinogen in humans (Salaspuro, 2003).And a recently published meta-analysis reported that, compared with nondrinkers, the pooled RR of GC was 1.07 (95%CI: 1.01-1.13)for alcohol drinkers and 1.20 (95%CI: 1.01-1.44)for heavy alcohol drinkers (≥4 drinks per day) (Tramacere et al., 2012).Turning to tea consumption, several animal experiments have suggested that green tea, which contains abundant polyphenols and catechins, specifically epigallocatechin-3-gallate (EGCG)5, might have a protective effect against cancers (Ahmad et al., 1999;Lambert, 2013;Yu et al., 2014).Gao et al's research showed that tea consumption was significantly associated with decreased risk of breast cancer in Chinese females, and the OR was 0.79 (95%CI: 0.65-0.95)(Gao et al., 2013).Zhou et al's systematic review revealed that green tea consumption was not associated with the risk of GC in both males and females with pooled odds ratio (OR) of 1.10 (95%CI: 0.76-1.60)and 0.99 (95%CI: 0.64-1.51)respectively (Zhou et al., 2008).
China has also witnessed numerous survey studies investigating the relationships between GC and tobacco, alcohol and tea consumption.Yet findings of these researches are seldom used by international researchers due to language and perhaps cultural barriers.This review aims at comprehensively summarizing the epidemiological evidences about the associations between risks of GC and tobacco smoking, alcohol drinking and tea intake derived from Chinese population in a hope to facilitate application, by international community, of research findings by Chinese researchers.

Data sources and search strategy
We utilized two approaches to locate as many relevant papers as possible.First, we searched the English literatures available by March 1th, 2014 via PubMed and Embase using the following search terms "(alcohol OR tea OR tobacco OR cigarette OR smoking) AND (gastric OR stomach) AND (cancer OR oncology OR tumor OR tumour OR malignan* OR carcinoma OR neoplasm*) AND (China OR Chinese)", where * represents wildcard characters.Meanwhile, we searched the Chinese literatures from China National Knowledge Infrastructure (CNKI) and China Biology Medicine (CBM) database via the same search terms.Second, we examined the bibliographies of relevant review papers for additional articles.This process was conducted iteratively until no new papers were identified.

Inclusion criteria
The inclusion criteria were paper: 1) written in English or Chinese; 2) belong to cohort or case-control (≥100 GC cases) study investigating the relationships between GC and behavioral factors that include tobacco smoking, alcohol or tea consumption; and 3) with ORs/ RRs and the corresponding 95%CIs or adequate original data for calculating them.

Data extraction
Extraction of descriptive data about the included studies utilized a data-extract form consisting first author, year of publication, study design, sample size, source of controls and adjusted OR/ RR with the corresponding 95%CI for each category (e.g.gender, subtype of exposure, duration of exposure, dose of exposure, anatomical subtype of GC) of tobacco smoking, alcohol and tea drinking.Two researchers performed the data extraction independently and discrepancies were solved by consensus.

Statistical analysis
Given that absolute risk of GC was low and that the RR in cohort studies approximated OR (Greenland, 1987), the study used OR to measure the associations of the three behavioral factors and GC risk.Pooled ORs with 95%CIs derived by pooling study-specific effect sizes using a random effects model.For dose-response analyses, studies included must provide the following data: number of GC case and non-case, median exposure (i.e.tobacco smoking, alcohol and tea drinking) dose and OR with 95%CI for each category of exposure.
If a study did not report the median exposure dose of a given category, we assigned a value for such dose by calculating the midpoint between the lowest and highest bound exposure doses.When the highest category was open-ended, we assumed the dose as 1.5 times the lowest bound of this category.Computation of trend from the correlated log OR estimates across categories of exposure employed a two-stage random-effects dose-response meta-analysis (Orsini et al., 2012) which took into account the between-study heterogeneity.More specifically, we established a restricted cubic spline model with four knots at the 5th, 35th, 65th, and 95th percentiles (Durrleman et al., 1989) of exposure using generalized least-square regression (GLST), taking into account the correlation between estimates for different expose categories to compute study-specific slopes (linear trends) (Orsini et al., 2006).P value for nonlinearity derived by testing the null hypothesis that the coefficient of the second spline equals zero (Greenland et al., 1992) and two-sided P value of less than 0.05 defined statistically sufficient to refuse the null hypothesis.All statistical analyses utilized software STATA version 12.0 (Stata Corporation, College Station, TX, USA).

Quality assessment
Assessment of methodological quality of the included studies used Newcastle-Ottawa Scale (NOS) (Wells et al., 2000).The scale provides a comprehensive score system with three broad aspects including selection of study groups, comparability between groups and ascertainment of exposure or outcome for both case-control and cohort studies.Total score came out from adding up the points awarded to each item.Only studies scored 7 or higher were considered as of high methodological quality.

Studies included
Total English articles retrieved from PubMed and Embase accounted for 523, of which 484 articles were excluded on the basis of title and abstract and another 18 were excluded after more detailed evaluation of fulltexts.Examination of reference lists identified additional 6 articles and 27 English papers finally met the inclusion criteria.Similarly, total Chinese papers retrieved from CNKI and CBM and papers excluded by title, abstract and full text screening were 1046 and 987 respectively.After adding 3 papers from reference tracking, a total of 62 Chinese articles met inclusion criteria.Putting papers published in both languages together, this review finally included a total of 89 articles.

Descriptive analyses
The 89 articles documented 5 cohort, 21 hospital-based case-control and 63 population-based case-control studies containing 25821 GC cases and 135298 non-cases.The sample sizes of case-control and cohort studies ranged from 202 to 2987 and 943 to 40508 respectively.The most frequently studied behaviors were tobacco smoking (n=74), followed by alcohol consumption (n=61) and tea drinking (n=33).The NOS scores of individual studies ranged from 6 to 9 with a mean score of 7.35.

Tea drinking and GC
Putting together, habitual tea drinking significantly decreased the risk of GC (overall pooled OR=0.67, 95%CI: 0.59-0.76);while dividing into subgroups by study design, the pooled effect was only statistically significant for the population-based case-control studies (OR=0.67,95%CI: 0.59-0.76)but not for the 5 hospital-based case-control studies (OR=0.67,95%CI: 0.28-1.26)(Figure 3).Tea consumption showed statistically significant associations with GC for all the subgroups categorized by duration of drinking, study quality, sample size, publication language and publication year (P<0.05)(Table 3).It demonstrated a 27% risk reduction for females (pooled OR=0.73, 95%CI: 0.55-0.98)but no significant effect for males (pooled OR=0.85, 95%CI: 0.66-1.09);a 46% risk reduction for non-cardia (pooled OR=0.54, 95%CI: 0.31-0.95)but not cardia GC.The result of green tea (pooled OR=0.62, 95%CI: 0.52-0.74)but other types of tea was consistent with that of the overall analysis.

Dose-response analyses
As shown in Figure 4, risk of GC increased stably as the dose of tobacco smoking rose from 0 to 30 cigarettes per day, while the risk decreased slightly when the dose was higher than 30 cigarettes a day.Similarly, the risk of GC increased sharply as the dose of alcohol drinking rose from 0 to 30 g/day; then the OR kept stable as the dose of alcohol drinking increased from 30 to 60 g/day; and when the dose became higher than 60 g/day, the OR began to increase again though at a much lower velocity.As for tea drinking, the OR remained stable at a relatively high level when the dose was less than 60 g/month; then the OR decreased consistently at a moderate speed as the dose increased.All the associations of GC with tobacco smoking, alcohol consumption, and tea drinking tested with statistical non-linearity.

Discussion
To our knowledge, this is the first review to summarize the associations between tobacco smoking, alcohol drinking, tea drinking and the risk of GC in Chinese population with literatures published in Chinese and English languages.Our findings indicated that tobacco smoking was associated with 1.62-fold higher risk of GC (95%CI: 1.50-1.74)compared with nonsmokers in Chinese population.This is consistent with a previous review based on international cohort studies, which reported a pooled RR of 1.53 (95%CI: 1.42-1.65)and a meta-analysis of international case-control studies with a pooled OR of 1.48 (95%CI: 1.28-1.71)(Ladeiras-Lopes R et al., 2008;La Torre et al., 2009).
Our review showed that alcohol drinking was associated with 1.57-fold risk of GC (95%CI: 1.41-1.76).This is a little lower than that of a previous reported pooled risk based on international case-control studies (pooled OR=1.77, 95%CI: 1.46-2.15)by Mahjub et al but apparently higher than the result documented in a recently published review which reported pooled RR of 1.07 (95%CI: 1.01-1.13)for alcohol drinkers and 1.20 (95%CI: 1.01-1.44)for heavy alcohol drinkers (Mahjub et al., 2007;Tramacere et al., 2012).These discrepancies may partly be attributed to differences in composition of studies reviewed by different authors.The majority (8 out of 11) of studies included in Mahjub and colleagues' review were hospital-based case-control studies; while Tramacere et al's review included as many as 16 cohort studies; and the bulk of our review consisted of population-based case-control studies.As our and other studies indicate (Table 1-3), ORs derived from hospital-based case-control studies tends to be higher than that from population-based case-control studies, which in turn tend to be higher than that from cohort studies.
Compared with previous reviews, our paper has several strengths: a) our analysis included almost all of the epidemiological studies carried out in Chinese population investigating the association between GC risk and tobacco, alcohol and tea consumption, involving as many as 25821 GC cases and 135298 non-cases; b) it comprised quite comprehensive subgroup analyses with 9 factors including study design, gender of subjects, anatomical subtype of GC, subtype of exposure, duration of exposure, study quality, sample size, publication language and publication year; and c) it included dose-response analyses for quantitatively assessing the non-linearity relationships between exposure dose and GC risk.Of course, our results should also be interpreted with caution for several limitations.First, the majority of the studies included were based on retrospective design which is prone to recall bias.Second, unpublished studies and papers published in languages other than English and Chinese, as well as articles did not provide ORs/RRs and the corresponding 95%CIs or adequate original data for calculating them were excluded in this review.Third, although major potential confounding factors had been adjusted in most studies, some unknown or residual factors could also result in exaggeration or underestimation of risk estimates.Last, the heterogeneity for the overall and part of the subgroup analysis were quite high (I 2 >50%) in our review and thus damages the validity of results.
The study revealed that tobacco smoking and alcohol drinking were associated with over 1/2 added risk of GC while tea drinking, about 1/3 lower risk of GC in Chinese population.Given that these three behaviors are very common in China, they merit special attention for related researchers and policy-makers.Future studies in this regard should:1) pursue larger sample size and more rigorous (e.g.cohort or population-based) studies with ample attention be paid to methodology quality; 2) emphasize the importance of adopting a consistent definition of exposures (e.g.alcohol drinking) and using uniform grouping criteria for exposure dose, duration and frequency so as to reduce the grouping bias and to determine the dose-effect relationships between them; 3) focus on identifying other factors that may affect the correction between tobacco, alcohol, tea and GC including behavioral factors (e.g.coffee consumption, dietary factors) and medical history (e.g.chronic atrophic gastritis, peptic ulcer); and 4) conduct more comprehensive and sophisticated analyses on the risk of GC to build multivariable risk prediction models (e.g.regression models, score systems).

Figure 3 .
Figure 3. Adjusted odds Ratios of Gastric Cancer for Tea Drinkers Versus Non Drinkers.C=cardia; N=noncardia; OR=odds ratio

Figure 4 .
Figure 4. Dose-Response Relations between Tobacco Smoking, Alcohol and Tea Drinking and Risk of Gastric Cancer.The plots were generated from randomeffects dose-response model; solid lines and the long dash lines represent the estimated odds ratio and its 95% confidence interval of the nonlinear relationship; short dash lines represent the linear relationship DOI:http://dx.doi.org/10.7314/APJCP.2014.15.20.8765Association of Risk of Gastric Cancer with Consumption of Tobacco, Alcohol And Tea in the Chinese Population

Table 1 . Summary Statistics for the Association between Tobacco Smoking and Gastric Cancer
Note: a single study may has more than one dataset; b ORs generated form random-effects analysis; c P value of Q test for heterogeneity test; NOS=Newcastle-Ottawa Scale; OR=odds ratio.

Table 2 . Summary Statistics for the Association between Alcohol Drinking and Gastric Cancer
Note: a single study may has more than one dataset; b ORs generated form random-effects analysis; c P value of Q test for heterogeneity test; NOS=Newcastle-Ottawa Scale; OR=odds ratio.

Table 3 . Summary statistics for the Association between Tea Drinking and Gastric Cancer
Note: a single study may has more than one dataset; b ORs generated form random-effects analysis; c P value of Q test for heterogeneity test; NOS=Newcastle-Ottawa Scale; OR=odds ratio.