Abstract
OBJECTIVE To reevaluate the effect of tobacco smoking on the risk of developing gastric cancer among the Chinese population.
METHODS Thirty articles from the literature both in Chinese and English from January, 1988 to present were identified and from which adjusted odd ratios (ORs) or relative risks (RRs) were combined by meta-analysis. Generalized least squares (GLS) for trend estimation of summarized dose-response data was carried out. All the analyses were performed using software of STATA version 10.0.
RESULTS Comparing current smokers with subjects who have never smoked, the summary effect values on gastric cancer with a 95% confidence interval (CI) were 1.67 (1.43-1.96) for case-control studies and 1.52 (1.17-1.96) for cohort studies, respectively. The combined effect values with a 95%CI for the comparison of smoking quantity of current smokers with the referent group were 1.41 (1.15-1.72) for case-control studies and 1.24 (1.02-1.52) for cohort studies, respectively. The combined effect values with a 95%CI for the comparison of accumulative years smoked of current smokers with the referent group were 1.07 (0.89-1.28) for case-control studies and 1.28 (0.95-1.74) for cohort studies, respectively. Dose-response meta-analysis showed that the gastric cancer risk will increase 50% (OR = 1.50, 95%CI: 1.34-1.67) with each 20 cigarett e per day increment and increase 14% (OR = 1.14, 95%CI: 1.10-1.18) with each 10 year increment of smoking.
CONCLUSION The gastric cancer risk among the Chinese population is significantly associated with tobacco smoking and the smoking quantity per day. Smoking cessation should be more strongly advocated.
keywords
Introduction
Gastric cancer is one of the leading cancers in China, with an estimation of 0.3 million deaths and 0.4 million new cases occurring in 2005[1]. Several factors were considered mainly responsible for the risk of developing gastric cancer, for example, low fruit and vegetable intake, high salt consumption[2,3], H. Pylori infection[4], and tobacco smoking[5,6]. Among these risk factors, tobacco smoking, as an avoidable lifestyle, has been considered the single most preventable one for the increased morbidity and mortality of disease, specifically for cancer, and remains a serious social and public health problem in China.
The surveillance of risky behaviors in 145 disease surveillance points (DSP) carried out in 2002 in China showed that the number of total smokers aged 15-69 years was about 350 million, 30 million more than that in 1996. Current smoking rates in the population as a whole were 31.4%, and in males and females 57.4% and 2.6%, respectively[7]. However, up until now, the results from studies focused on the association between tobacco smoking and gastric cancer risk have varied. With 1/5 of the world’s population, the incidence and mortality of gastric cancer in China plays a very important role in determining the cancer burden at a global level. So it is necessary to comprehensively reevaluate estimations of the gastric cancer risk associated with smoking in the Chinese population by systematic review of epidemiological evidence. Data provided by the present study is expected to encourage the public health policymakers in China to make efforts to promote smoking cessation.
Materials and Methods
Source of data
A search of MEDLINE, PUBMED, HIGHWIRE PRESS for English and CNKI, CHONG-QING VIP database for Chinese articles from January, 1988 until the present was conducted using the following terms: ‘tobacco’ ‘cigarette’, ‘lifestyle factor’, ‘gastric cancer’, ‘stomach cancer’, ‘cohort study’, ‘case-control study’, ‘China’ and ‘Chinese’ or corresponding Chinese words. When necessary, manual searches of references from relevant articles were performed. Two researchers independently screened the list of references and excluded inappropriate papers. Disagreements were discussed with another reviewer and resolved by consensus. In addition, in order to perform the dose-response meta-analysis, our search required that the involved studies contain similarly classified items and provide the ORs or RRs which were adjusted for multiple potential confounding factors in the logistic regression model. Meanwhile, papers of poor quality, repeated reports, those with little useful information and those with obviously discrepant classification were excluded by assessment. When gender-specific or histologic-specific estimates were available, they were considered separately as if obtained from different studies.
Data management
Meta-analysis was conducted to summarize the ORs or RRs, which were provided by the papers and were focused on examining the gastric cancer incidence or mortality risk associated with cigarette smoking. These indexes were also estimated using different definitions of smoking status or different classifications as follows: i) current smokers or those who have ever smoked vs. those who have never smoked; ii) smoking quantity for current smokers (preference was given to those who smoked closest to 20 cigarettes per day if this information was provided, and this value was then used in the dose-meta analysis) vs. the referent group (those who have never smoked or smokers who smoke the least). iii) accumulative smoking years for current smokers (preference was given to those who smoked closest to 10 years if this information was given) vs. the referent group (those who have never smoked or smokers with the least accumulative smoke years).
The combined OR or RR was the average of the logarithm of the original ORs or RRs which was weighted by the inverse of its variance. The random effect model or the fixed effect model was used according to the heterogeneity of the test results. Publication bias was examined through the Egger’s regression asymmetry test[8]. Generalized least squares (GLS) for trend estimation of summarized dose-response data was carried out[9]. This method calculated the correlation between risk estimates for separate exposure levels depending on the same reference group. The summary estimate was the pooled coefficient b in the linear-logistic regression model lnOR/RR = bX, where X stands for the difference between each group and the reference group. The individual slopes of each study were combined by weighted average, using the inverse of their variances as weights. In our study, the average consumption should be rescaled as follows: if the median or mean consumption per group was not presented, we assigned the midpoint of the upper and lower boundaries as the average; if the upper boundary of the highest group was not provided, we assumed that it had the same amplitude as the previous group and when the lowest group was open-ended, the lowest boundary was assumed to be zero.
Statistical analysis
All statistical analyses were conducted using STATA version 10.0 and all tests of statistical significance were two-sided. The default test level α was 0.05 if not specified.
Results
Main feature of studies
At first, 33 articles were included in our systematic review after searching the literature, but articles with represented repeated reports were excluded by assessment. Thus the article written by Sun et al.[10] in 2000 was deleted because it did not describe the adjusted covariant; therefore, the article published in 1999[11] was retained. An article written by Bao et al.[12] in 2001 was kept in our study because it was a case-control study specifically addressing the association with smoking, alcohol consumption and stomach cancer, but a paper from 2003[13] which focused on some other unrelated risk factors was not used. The paper by Li et al.[14] was almost the same study as a paper of Wang et al.[15] published in 2007, but we kept the latter one as it was published more recently. According to the criterion we detailed above, at least 30 articles were ultimately included in our systematic review. Of all the papers, 15 were case-control studies in Chinese[11,12,15-27], 2 were cohort studies in Chinese[28,29], 7 were case-control studies in English[30-36], and 6 were cohort studies in English[37-42]. Furthermore, only 2 studies[37,38] were published before 1998 and the others after 1998 (including 1998). In addition, 8 studies[12,21,30,37,39-42] reported the results related to the subjects’ sex and only 4 studies[30,35,36,42] provided information regarding histologic type of gastric cancer.
Outcomes
Case-control studies
Comparing current smokers or those who have ever smoked vs. those who have never smoked
The summary OR was 1.67 (95%CI: 1.43-1.96, Q = 50.46, df = 20, P <0.001) for the case-control studies. The single OR from each study is shown in Figure 1. When case-control studies were separated by language type, the combined ORs were 1.80 (95%CI: 1.44-2.26, Q = 23.211, df = 10, P = 0.010) for studies published in Chinese and 1.57 (95%CI: 1.24-1.97, Q = 25.722, df = 9, P = 0.002 ) for studies in English.
Comparing smoking quantity of current smokers with the referent group
The combined OR was 1.41 (95%CI: 1.15-1.72, Q = 39.52, df = 16, P = 0.001) for case-control studies. The single OR from each study is shown in Fig. 2. When case-control studies were separated by language, the combined ORs were 1.86 (95%CI: 1.33-2.60, Q = 20.65, df = 8, P = 0.008) for studies in Chinese and 1.11 (95%CI: 0.93-1.32, Q = 7.286, df = 7, P = 0.400) in English.
Comparing accumulative years of smoking of current smokers with the referent group
Comparing current smokers whose average accumulative years of smoking was close to 10 years with the referent group, we obtained a summary OR of 1.07 (95%CI: 0.89-1.28, Q = 6.31, df = 10, P = 0.789. The single OR of each study is shown in Figure 3. The combined ORs were 0.98 (95% CI: 0.64-1.49, Q = 3.779, df = 3, P = 0.286) for case-control studies in Chinese and 1.09 (95%CI: 0.89-1.34, Q = 2.315, df = 6, P = 0.889) in English.
Cohort studies
Comparing current or those who have ever smoked vs. those who have never smoked
Further analysis showed that the combined RR was 1.52 (95%CI: 1.17-1.96, Q = 20.92, df = 6, P = 0.002) for the cohort studies. The single RR of each study is shown in Fig. 4. The combined RR for cohort studies in English was 1.34 (95%CI: 1.09-1.66, Q = 11.308, df = 5, P = 0.046). The combined RR for cohort studies in Chinese could not be calculated because of only one relative study was available[25].
Comparing smoking quantity of current smokers with the referent group
The summary RR was 1.24 (95%CI: 1.02-1.52, Q = 8.04, df = 6, P = 0.235) for cohort studies when comparing the quantity of 20 cigarettes smoked per day with referent group. The single RR of each study is shown in Fig. 5. As there were no cohort studies in Chinese, we estimated that the combined RR for cohort studies in English was 1.24 (95%CI: 1.02-1.52, Q = 8.04, df = 6, P = 0.235).
Comparing accumulative years of smoking of current smokers with the referent group
For current smokers whose accumulative years of smoking was close to 10 years, the summary RR estimate was 1.28 (95%CI: 0.95-1.74, Q = 13.79, df = 6, P = 0.032) compared with the referent group. The single RR of each study is shown in Fig. 6. Because there was only one cohort study[26] in Chinese, we estimated that the combined RR for cohort studies in English was 1.07 (95%CI: 0.86-1.33, Q = 7.295, df = 5, P = 0.200).
The combined OR/RR for the above studies is shown in Fig. 7. We can conclude through reviewing the results that the gastric cancer risk among the Chinese population is significantly associated with tobacco smoking and a greater quantity of tobacco per day, but the relationship between the accumulative years of smoking and gastric cancer risk is not certain.
Publication bias in analysis
The publication bias was examined through the Egger’s regression asymmetry test (Table 1). A 0.1 level of significance was used in the statistical tests to improve the sensitivity of detecting publication bias. We could see that publication bias may exist in both the case-control and cohort studies between the quantity of tobacco smoked and the cohort studies on the relationship of tobacco smoking or not with gastric cancer risk.
Dose-response meta-analysis
To use the GLS method, the frequency distribution of case and control subjects for case-control studies or the incidence rate data for cohort studies and OR/RR with variance estimates for at least 3 quantitative exposure categories which have the same reference category were required. Ultimately, only 7 case-control studies[11,21,25,30,33,34,36] met the required standards in our study of dose-response meta-analysis associated with the relationship between quantity of tobacco smoked and gastric cancer. The estimated summary OR of gastric cancer for an increase in cigarette consumption of 20 cigarettes per day was 1.50 with 95% CI: 1.34-1.67. According to the statistics (Q = 30.19, P = 0.31), the linear model was used and the variable number of cigarettes consumed per day (β = 0.02019, P < 0.001) was included in this model.
The next dose-response meta-analysis performed was to calculate the relationship between accumulative years of smoking and gastric cancer among 6 case-control studies[11,21,25,30,33,36]. The estimated summary OR of stomach cancer for 10 accumulative years was 1.14 (95%CI: 1.10-1.18). The goodness-of-fit test detected no further problems with the fitted model (Q = 25.53, P =0.43) and included the variable accumulative years of smoking (β = 0.01297, P < 0.001) in the model.
Discussion
Many, though not a11 of the epidemiological studies focused on the Chinese population showed that smoking is associated with gastric cancer. To the best of our knowledge, this meta-analysis is the most comprehensive study addressing the association between cigarette smoking and gastric cancer among the Chinese population. A significant positive association was found between tobacco smoking and gastric cancer in our study, which was consistent with the findings of other meta-analysis studies[43,44].
Although the carcinogenic effect of tobacco smoking on gastric mucosa has been demonstrated in vitro[45] and in experimental models[46], the exact mechanisms by which tobacco smoking affects the development of gastric cancer remain unclear. Nitrosamines form a large group of chemical carcinogens, due to the presence of the N-nitroso group, and are a significant factor regarding the etiology of gastric cancer[47]. A number of studies have demonstrated that cigarette smoking increases the frequency of the transition of gastric mucosa to precancerous lesions[48]. Further, smoking-related DNA adduct levels are significantly higher in cases of gastric cancer among smokers[49]. This evidence implies that tobacco smoking may be considered an important risk factor for gastric cancer.
Different from the previous studies which have been published and which focused on the relationship between tobacco smoking and gastric cancer risk among the Chinese population, our analysis included 8 cohort studies[28,29,37-42]. In addition, our study covered 3 classifications of tobacco smoking status. GLS regression was also first used to analyze the dose-response relationship among the Chinese population. Moreover, the OR or RR we used adjusted for various potential confounders, such as age, education, occupation, and the confounders from the multivariate logistic regression model of each separate study, therefore, made the summary effect estimation more reasonable and reliable.
Several reasons could explain the publication bias which existed in studies of tobacco smoking quantity and in cohort studies on the relationship of tobacco smoking or not smoking with gastric cancer risk. First, although defined ORs or RRs were used for the comparison of the group composed of smokers smoking close to 20 cigarettes per day, with accumulative years of smoking close to 10 in the referent group, some incorrect classification of the subjects was inevitable, resulting in recall or reporting bias. Secondly, studies with positive associations may be more prone to being published than studies with negative results. Finally, the language the papers were written in constitute another bias factor. For example, in the case-control studies on the relationship between quantity of tobacco smoked and gastric cancer risk, the combined ORs were 1.86 (95%CI: 1.33-2.60, Q = 20.65, df = 8, P = 0.008) for studies in Chinese and 1.11 (95%CI: 0.93-1.32, Q = 7.286, df = 7, P = 0.400) in English. Typically authors who write good quality papers, especially cohort studies in China, like to have their papers published in foreign academic journals in English which can reach more readers throughout the world; this may introduce some publication bias. For instance, there were only 2 papers[28,29] written in Chinese included from the 8 cohort studies in our study.
Based on constructing an approximate covariance estimate for the log relative risk and estimating a corrected linear trend using GLS method, trend analysis of multiple studies were conducted[50]. Because the ORs we collected had been adjusted in the original literature, the two goodness-of-fit tests (Q = 30.19, P = 0.31 and Q = 25.53, P = 0.43) did not detect problems with the linear-log model; therefore, there was no need for dose-meta analysis to test potential sources of heterogeneity.
In summary, this meta-analysis supported the conclusion that tobacco smoking may be associated with an increased risk of gastric cancer. About 50% of the risk of developing gastric cancer will decline if the smoker decreases the smoking quantity by one pack of cigarettes (20 cigarettes) per day, and 14% of the risk will be reduced for each 10 years of cessation if one smoker quits smoking. Studies in the future should include more studies from literature searches, particularly, articles in English, and should be controlled for more potential confounders. Whether the association differs according to histologic subtype or anatomic subsite should be included in the future studies also.
Footnotes
This work was supported by a grant from the National Natural Science Foundation of China (No.30872176).
- Received June 1, 2009.
- Accepted August 2, 2009.
- Copyright © 2009 by Tianjin Medical University Cancer Institute & Hospital and Springer