Physical activity and risk of gastric cancer: a meta-analysis of observational studies
- 1Department of Global Health and Population, Harvard School of Public Health, Boston, Massachusetts, USA
- 2Newgate Medical Services Limited, Ikorodu, Lagos, Nigeria
- 3Community Health Unit, Lagos State General Hospital, Igando, Lagos, Nigeria
- 4Department of Surgery, Lagos State University College of Medicine, Ikeja, Lagos, Nigeria
- Correspondence to Dr Nasiru Akanmu Ibrahim, Department of Surgery, Lagos State University College of Medicine, 1-5 Oba Akinjobi Way, PMB 21266, Ikeja, Lagos, Nigeria;
- Received 22 October 2013
- Revised 2 December 2013
- Accepted 19 December 2013
- Published Online First 16 January 2014
Background Studies evaluating the relationship of physical activity and stomach cancer risk have yielded inconsistent and largely inconclusive results. We therefore conducted a systematic review and meta-analysis of observational studies that assessed the relationship between physical activity and risk of gastric cancer.
Methods Following a standard protocol, we searched medical literature databases (PubMed, EMBASE, CINAHL, PsycINFO and Google Scholar) from inception to July 2012, and conducted a random effects meta-analysis.
Results Seven prospective cohorts and four case–control studies of physical activity and gastric cancer risk, with 1 535 006 people and 7944 cases of gastric cancer were included. We found a modest protective association between sufficient physical activity and gastric cancer risk (relative risk: 0.81 (95% CI 0.69 to 0.96); I2=68.5%) in the prospective studies and (relative risk: 0.78 (95% CI 0.66 to 0.91); I2=0%) in case–control studies. The association appeared weaker in smokers than in non-smokers (p heterogeneity=0.035). The association may also be weaker for gastric cardia cancer relative to the distal non-cardia subtypes. Physical activity type (recreational or occupational), intake of alcohol, total energy intake, consumption of fruits and vegetables and infection with Helicobacter pylori had no influence on the association. The effect measure from cohort studies (relative risk: 0.82 (95% CI 0.70 to 0.97); I2=61.7%) and case–control studies (relative risk: 0.83 (95% CI 0.66 to 1.04); I2=49.8%) did not differ materially at higher physical activity levels.
Conclusions We conclude that a regular physical activity may be protective against stomach cancer risk.
The burden of non-communicable diseases is on a steady rise globally, especially in developing countries.1 Gastric cancer is the fourth most common cancer globally and the second leading cause of cancer-related mortality worldwide (10.3/100 000 people).2 The disease is more common in men,2 and prognosis is extremely poor.3 More than 70% of cases occur in developing countries, particularly in Eastern Asia.2
Several studies have sought to elucidate the aetiology and risk factors of gastric cancer. Infection with Helicobacter pylori,3 ,4 lifestyle factors, such as diet,3 ,5 smoking6 and excessive alcohol consumption,7 and premalignant lesions8 are established risk factors of the disease. Physical inactivity, a potentially modifiable lifestyle factor, has been associated with an increased incidence of cancers of many body sites.9 ,10 Although many epidemiological studies have evaluated the relationship between physical activity (PA) and cancer of the stomach, their findings have been inconsistent and largely inconclusive. Some studies have suggested that the association is modified by sex11 and region of the stomach involved.12
No prior rigorous systematic review or meta-analysis has been conducted to synthesise the evidence from the existing epidemiological studies in this regard. Therefore, we conducted a systematic review and meta-analysis to quantitatively evaluate the association of PA with gastric malignancies and assess factors contributing to inconsistency across the studies.
We aimed to identify and examine the studies evaluating the relationship between PA and risk of gastric cancer. We followed the meta-analysis of observational studies in epidemiology (MOOSE) guidelines in the design, analysis and reporting of this study.13
Search strategy and selection of studies
In accordance with a prespecified study protocol, we searched the medical literature databases of PubMed, EMBASE, PsycINFO, CINAHL and Google Scholar for original research articles published as of July 2012. Medical Subject Headings (MeSH) and keywords representing gastric cancer and PA were employed. We searched PubMed using the following query: (("Motor Activity"[Mesh] OR "physical activity"[All Fields] OR "exercise"[All Fields] OR "walk"[All Fields]) AND ("stomach"[MeSH Terms] OR "stomach"[All Fields] OR "gastric"[All Fields]) AND ("cancer"[All Fields] OR "neoplasm"[All Fields])). We employed a similar search strategy for EMBASE, using the following query: 'motor activity'/exp OR 'physical activity'/exp OR 'exercise'/exp AND 'stomach'/exp AND 'neoplasm'/exp. In CINAHL and PsycINFO, we used: ‘physical activity’ AND ‘stomach’ AND ‘cancer’. No restrictions by age, year of publication or language were applied, though all the citations we found were available in English. We also reviewed the reference list of included articles but found no additional papers. Two authors (AIA-1 and MOO) independently screened the titles and abstracts for eligibility and examined the full texts of selected articles. The inclusion criteria included availability of estimates of the association between PA and risk of stomach cancer. Cross-sectional studies and case series were excluded. There was 95.5% inter-rater agreement for eligibility, and discrepancies were resolved by discussion among the authors.
Data extraction from full-text articles was carried out using a standardised comprehensive extraction sheet. For each eligible study, data on study design, population, follow-up, outcomes, covariates and findings were extracted. Specific variables extracted include ethnicity/race, country, gender, age, sample size, body mass index (BMI), duration of follow-up, measurement of PA, outcome ascertainment, information on dropout and response rates and information about potential confounders.
We assessed the studies using the Newcastle Ottawa Scale for nonrandomised studies, which evaluates studies based on selection, comparability, exposure assessment (for case–control studies) and outcome ascertainment (for cohort studies).14 Two investigators (AIA-1 and MOO) independently assessed the studies and disagreements were resolved by consensus. Quality scores were not a basis for exclusion. Individual factors such as objective exposure ascertainment, repeated measurements and control for confounding due to dietary factors and alcohol consumption were employed in meta-regression to explore the heterogeneity. All studies included in the main meta-analyses adjusted for age, BMI and smoking. To varying extents, the studies also adjusted for sex,15–17 alcohol intake,12 ,16 ,18–20 H pylori infection,16 dietary factors, such as intake of total calories,16 ,18 red meat,11 ,12 ,16 salt,19 fruits16 and vegetables,11 ,12 ,21 ,22 occupation19 and diabetes.18
We considered the influence of PA on the risk of gastric cancer in our primary analysis, and the influence in subgroups determined by subtypes of gastric cancer and gender. We employed guidelines from the WHO’s contemporary recommendations for PA in adults,23 a measure of duration and intensity of PA. A person is ‘sufficiently active’ if they engage in 150 min of moderate intensity aerobic PA throughout the week or 75 min of vigorous intensity PA or an equivalent combination of both.23 We classified a person as ‘highly active’ if they engage in 300 min of moderate intensity exercise or 150 min of vigorous intensity exercise per week.23 The ‘inactive’ or ‘insufficiently active’ category included persons whose reported PA was less than that required to meet the sufficiently active category. In selecting appropriate effect estimates to pool, we considered categories that best fit the guidelines and preserved the relative ranking of the categories presented in the primary articles. For the adjusted analysis, we chose effect measures with the greatest degree of control for confounding. If a study reported multiple measures, we selected the one identified by the authors as the primary adjusted model. If a study reported only distinct effect measures for men and women, or for the subtypes of gastric cancer, we included them separately in the primary analysis. We requested additional data for the subgroup analyses from the authors of eight primary articles but only one group provided the needed data.16 No conversion was carried out for ORs extracted from the case–control studies since gastric cancers are relatively rare events. However, we, considered hazard and ORs in separate models. For the unadjusted analysis, we calculated risk and ORs using available information reported in the selected articles or provided by the authors.
Pooled effect estimates and 95% CIs were obtained using random-effects model which allows for variability of estimates across populations. Forest plots (with CI) were constructed to demonstrate the individual and pooled effect estimates, with their CIs, and to allow visual inspection for study heterogeneity. Heterogeneity across studies was formally assessed with the Q and I2 statistics and explored using meta-regression. We explored heterogeneity due to study-specific characteristics specified a priori based on empirical relationships with exposure and outcome: BMI, age, sex, smoking, dietary factors.3 ,5 ,6 ,24 We also considered year of study, duration of follow-up and type of PA measured for practical reasons. To evaluate heterogeneity due to age at baseline, we used mean ages, median ages or the midpoint of a specified reported age range. Publication bias was evaluated with Begg's and Egger's tests. Influence analysis was conducted to obtain pooled estimates after omitting each individual study estimate. Significance was taken as p<0.05. All statistical analyses were carried out in Stata V.11 (College Station, Texas, USA).
Our preliminary search identified 350 citations. After excluding 35 duplicates and 283 other studies that did not meet the inclusion criteria, we extracted data from 11 studies for the main meta-analysis: 7 prospective cohorts11 ,12 ,16 ,18–20 ,25 and 4 case–control studies15 ,17 ,21 ,22 (figure 1). Combined, these studies evaluated 1 535 006 people and 7944 cases of gastric cancer, in more than 10 countries across Europe,16 ,20 ,25 Asia11 ,18 ,22 and North America.12 ,15 ,17 ,19 ,21 Study participants were predominantly men (63%), with mean age across studies varying from 49 to 62 years. Four studies reported information for multiple PA types,16 ,21 ,22 ,25 two studies collected information about overall PA,12 ,16 ,18 ,21 ,25 four studies for leisure-time PA alone11 ,15 ,19 ,20 ,22 and one for occupational PA alone.17 Exposure assessment varied, based on validated11 ,12 ,16 ,18 ,20 ,25 or non-validated questionnaires.19 Outcome information was largely derived from cancer registries using diagnosis made by managing physicians and pathologists11 ,12 ,16 ,18–20 ,25; and two studies also used death records18 ,20 as the source of outcome information. Three studies restricted cases to primary adenocarcinoma of the stomach.16 ,17 ,19 Quality scores ranged from 5 to 8, median score of 7 of a maximum possible of 10 (see online supplementary table S1).
We evaluated the risk of gastric cancer among those who were sufficiently active compared with those who did not meet the study definition of sufficient PA. First, we pooled adjusted estimates by study design (7 cohort and 4 case–control studies) in separate models and obtained relative risks (RR) of 0.81 (95% CI 0.68 to 0.96) and 0.78 (0.66 to 0.91), respectively (figure 2). The protective association was reduced in unadjusted analyses as shown by pooled RRs that were closer to null (RR=0.84 (0.71 to 0.98) from the cohort studies, and 0.85 (0.66 to 1.09) from the case–control studies). There was a significant heterogeneity in the pooled adjusted (I2=61.7% (p=0.007)) and unadjusted (I2=85.4% (p<0.001)) RR estimates from cohort studies (table 1). There was also a significant heterogeneity in the unadjusted RR from case–control studies (I2=76.1% (p=0.002)) but not in the adjusted RR (I2=0.0% (p=0.445)). Meta-regression analysis suggests that heterogeneity in the adjusted RR estimates from cohort studies was partly attributable to a history of smoking (p heterogeneity=0.035). The association was four times weaker in smokers than in non-smokers. Heterogeneity due to sex (p=0.055) and BMI (p=0.05) was near significant. The association appeared to be 1% weaker in men than in women and 10% weaker as BMI increases. We also considered H pylori infection in meta-regression using national H pylori infection rates obtained from external sources26 ,27 but found no influence on heterogeneity (p=0.642).
We considered the effect of higher levels of PA on the risk of gastric cancer, and the RRs did not differ materially from those with sufficient PA (figure 3). The pooled RR from adjusted HRs for the seven studies was 0.82 (0.70 to 0.97), with a marked heterogeneity (I2=68.4% (p=0.001)), and 0.83 (0.66 to 1.04) for the four case–control studies with no significant heterogeneity (I2=49.8% (p=0.093)). The pooled unadjusted RR was 0.74 (0.60 to 1.24) for cohort studies (I2=82.4% (p<0.001)) and 0.78 (0.59 to 0.91) for case–control studies (I2=85.7% (p<0.001)). Heterogeneity in the unadjusted estimates from cohort studies was also partly influenced by a history of smoking (p=0.011).
We considered the risk of cancer in the different anatomical regions of the stomach but were unable to conduct a meta-analysis because the studies were too few. Only three studies provided estimates for the influence of PA on the risk of gastric cancer, by the anatomical region involved.12 ,16 ,19 The individual HRs for non-cardia cancer ranged from 0.50 to 0.62, and were significant. The HRs for gastric cardia cancer ranged from 0.71 to 0.93, but were not significant. The ORs from case–control studies did not, however, had a consistent pattern.
Publication bias and sensitivity analysis
Influence analysis did not suggest that the pooled effect estimates of sufficient PA were dominated by any of the individual studies and omission of any of the studies did not change the effect estimate significantly (results not shown). There was no graphical (figures 4 and 5) or statistical evidence of publication bias in the studies of the effect of sufficient PA among cohort studies (p=0.661) and case–control studies (p=0.480) on the risk of gastric cancer based on Egger's test.
In sensitivity analysis (table 2), we evaluated alternative approaches to the main meta-analysis: by pooling case–control and cohort studies for greater statistical power, excluding Yun et al11 and Wen22 with relatively low-quality score of 5,11 ,22 including the estimates from the national sample from Campbell et al15 for recent PA in the past 2 years only and replacing estimates from Vigen et al17 with an average annual activity index before 65 years. These alternative approaches did not alter pooled risk estimates from our main analysis remarkably.
These meta-analyses of observational studies evaluated the influence of PA on gastric cancer risk and found that sufficient PA may provide modest protection against gastric cancer. We obtained consistent pooled results from cohorts and case–control studies, and the meta-analyses were robust to sensitivity analysis. Primary studies included in the meta-analyses either reported a clear inverse association11 ,12 ,15 ,16 ,19 or found no association.17 ,18 ,20–22 ,25 Our meta-analyses suggest a 19% protective association of sufficient PA on gastric cancer risk.
Physical inactivity is an important risk factor for chronic morbidity and disability.28 Prevalence of physical inactivity is very high globally, and greater than 10% of cases of certain cancers are attributable to physical inactivity.28 PA likely protects against chronic diseases such as gastric cancer by preventing chronic inflammation through upregulation of antioxidative defence systems29 and by reducing the levels of potentially carcinogenic hormones such as insulin-like growth factor30 and leptin.31
Findings from our subgroup analyses may be explained by lifestyle factors. The meta-regression analyses suggest that tobacco smoking may modify the relationship of PA and gastric cancer, reducing the magnitude of the protective association. Although there is conclusive evidence that smoking increases the risk of gastric cancer, the pathogenetic mechanisms are uncertain.32 We postulate that effect modification by tobacco smoking is induced via inflammatory processes33 that probably counteract the protective influence of PA on chronic inflammation. Further research may be able to investigate the existence of interactions between smoking and PA in the aetiology of gastric cancer.
Heterogeneity due to sex and BMI was near significant. Our finding of a slightly weaker association in men than in women seems biologically plausible since lifestyle habits (including smoking and alcohol consumption) differ by sex. Furthermore, the benefits from regular PA may also be modulated by hormonal conditions, and PA questionnaires may not be adequately adapted to satisfactorily measure the traditional patterns of PA in women.18 Only Wen22 reported the influence of domestic PA on gastric cancer risk and the type of PA did not influence heterogeneity in our subgroup analyses. Previous studies have shown that BMI may be associated with gastric cardia cancer but not with non-cardia cancer.3
The heterogeneity due to selected factors would reflect population characteristics and availability of information in each study. First, though infection with H pylori is a strong risk factor for gastric cancer,4 only one study provided information for exploring this further.16 Using national infection rates as surrogates in meta-regression, we did not find H pylori infection to be significant determinants of heterogeneity. Second, dietary assessment and adjustment for confounding due to diet were inadequate in the primary studies. There is evidence that excess intake of dietary salt34 and processed meat35 predisposes to gastric cancer while intake of dietary fibre,36 fruits and vegetables37 is protective. Third, it is known that excess alcohol consumption increases the risk of gastric cancer.7 However, only Huerta et al16 provided estimates of alcohol consumption by participants. We were unable to satisfactorily explore these factors in meta-regression due to inadequate information.
Estimates from the cohort studies suggest that the influence of PA on gastric cancer appears to differ by anatomical region of the stomach. The protective association seems strong in the distal non-cardia regions and weak in the proximal cardia of the stomach. These findings may be explained by the epidemiology and pathogenesis of the two anatomical subtypes. Gastric cardia cancer is five times more common in men than in women.3 While stomach cardia cancer shares risk factors such as gastro-oesophageal reflux disease, Barrett’s oesophagitis and obesity with oesophageal malignancies, H pylori and dietary factors contribute to non-cardia cancer risk.3
Exposure assessment is a potential source of error in research. Most studies reported PA as a composite of duration, intensity and frequency, using questionnaires. We did not separately evaluate the different domains of duration, intensity or frequency because very few studies reported these. A true biological lagging of the effect of PA on cancer development is likely though most included studies did not evaluate or consider this effect. In one cohort study, each participant contributed person-time 3 years after PA assessments were conducted and found very strong protective associations of moderate-to-vigorous PA with gastric cancer risk (RR=0.5).19 Another study excluded cases diagnosed within the first 2 years of follow-up.16 Potential misclassification in exposure assessment may have occurred since the studies evaluated PA using different validated and non-validated questionnaires; such misclassification is likely to be random and unrelated to outcome ascertainment, and would therefore not explain our positive findings of the inverse association.
Our database search yielded 11 eligible case–control and cohort studies. While repeated measures may better reflect the true exposure of subjects over time and better predict the outcome, only one study measured PA more than once.16 Biased selection of participants and recall of exposure information may have occurred in the case–control studies but this did not influence our results. Finally, the studies were conducted in Europe, Asia and North America, and the results cannot be generalised to other regions.
In summary, our meta-analysis indicates that regular PA may be protective against incidence of stomach cancer. This has important implications for aetiology and prevention of gastric cancer and additional evidence that an overall healthy lifestyle and regular PA would be beneficial in the prevention of non-communicable diseases.
What are the main findings?
Physical activity (PA) may protect against gastric cancer risk.
The protective association appears to be weaker in smokers than in non-smokers.
The protective association appears to be slightly weaker in men than in women at the highest levels of PA.
The protective association may also be weaker for gastric cardia subtypes compared with the distal non-cardia types, but this requires further evaluation.
How it might impact on clinical findings?
This article sheds more light on the protective role of PA against gastric cancer.
This may strengthen health-workers’ motivation for exercise referral and smoking cessation interventions, especially in populations with high cancer burden.
Efforts to control non-communicable diseases should include a comprehensive plan to promote PA.
The authors are grateful to Dr Fumiaki Imamura (MRC Epidemiology) and Dr Moshood Omotayo (Cornell) for their valuable comments during preparation of the manuscript.
Contributors NAI, MOO, AIA-1 and AIA-2 contributed to the conception and design of the study. AIA-1 and MOO conducted the search, screened the articles and extracted the data. AIA-1 conducted the analysis and wrote the first draft of the article. MOO and AIA-2 provided comments on the manuscript. NAI oversaw the whole process and is the study guarantor. All authors read and approved the final manuscript.
Competing interests None.
Provenance and peer review Not commissioned; externally peer reviewed.