Introduction

The overall incidence of stomach adenocarcinomas, which represent 95% of total gastric tumours, has been steadily declining for several decades [1]. Nevertheless, gastric cancer still remains as the fourth most common type of cancer worldwide with 934,000 estimated new cases per year [2]. Furthermore, whereas this declining trend regards the incidence of the most common distal (i.e., non-cardia) forms of gastric adenocarcinomas (GAC), tumours of the proximal region of the stomach (cardia) and those of the gastro-oesophageal junction (GEJ) are increasing in high-income countries [3, 4]. Oesophageal cancer appears to follow a similar pattern, with overall figures falling but drastic increases in incident oesophageal adenocarcinomas (OAC) in Europe and the United States [3–5].

This common rising pattern of adenocarcinomas of the oesophagus and proximal stomach in Western affluent areas might suggest the existence of a shared environmental background that would account for a common aetiology of these less frequent, albeit increasing, types of cancer. Among those lifestyle factors that have been studied in relation to the risk of cancer, there is physical activity (PA). Several tumours, especially those of the colon, breast, lung and endometrium have been shown to be associated with –and, to a variable extent, preventable by– PA to date [6–10]. But the association in relation to gastric or oesophageal cancer has been less studied, and the evidence for a protective role of PA is not yet conclusive [6, 11–17]. Furthermore, most of the studies did not show results by site or histological type of the tumours. According to recently published prospective data, the putative protection of PA on gastric cancer might be dependent on tumour site [18] and sex [19, 20]. The evidence for an association between PA and oesophageal cancer is even more scarce [15, 16, 20].

Since previous evidence is limited, our aim was to analyse the association of PA with GAC and OAC within EPIC (European Prospective Investigation into Cancer and Nutrition), a large-scale prospective study conducted in over half a million participants across ten European countries.

Subjects and methods

EPIC is an ongoing multi-centre prospective study carried out in 521,448 participants, mostly aged 25–70, recruited in 23 centres from 10 European countries between 1992 and 2000. The main aim of the EPIC study was the investigation of lifestyle (including diet), metabolic and genetic determinants of cancer and other chronic diseases, with a large-scale prospective epidemiological setting. Extensive dietary information of the previous year was gathered by means of validated questionnaires. Anthropometrical and lifestyle data were collected at recruitment, as well as information related to health issues, such as history of previous illness (cancer, diabetes, cardiovascular problems), or surgical operations. Blood samples were collected from approximately 74% of the EPIC participants. The rationale and detailed methodological issues of the study have been addressed in full in previous publications [21, 22].

Subjects were recruited from the general adult population residing in a given town or geographical area. In some centres (France, Utrecht—the Netherlands and Naples—Italy), only women were invited to participate, whereas half of the Oxford cohort consisted of non-meat eaters, and a large proportion of participants from Spain and the Italian centres Ragusa and Turin were blood donors. All participants gave their written informed consent. The protocols of the study were approved by the Ethical Review boards of the International Agency for Research on Cancer (IARC, Lyon) and the respective national/local Ethical Committees.

Prevalent cancer cases of any site according to baseline questionnaires were excluded for the present analysis (n = 23,633). Further exclusions included subjects lost to follow-up (n = 2,041), those lacking dietary information (n = 6,839) and participants for whom PA data were not available, including the whole cohorts from Norway (n = 37,725) and Umeå (n = 13,297). Eight participants initially diagnosed with cancer but not confirmed as cancer cases after examination were excluded for analyses. The final study sample consisted of 420,449 participants (130,087 men and 290,362 women) from 9 European countries: Denmark, France, Germany, Greece, Italy, Spain, Sweden, the Netherlands and United Kingdom).

Exposure assessment

Occupational activity

Participants were asked to choose the category that most accurately described the physical demand of their job, out of four possible options: sedentary occupation (e.g., office work), standing occupation (e.g., hairdresser, shop assistant, security guard), manual work (e.g., plumber, electrician, carpenter, cleaner, nurse) and heavy manual work (e.g., docker, bricklayer, miner). Due to small numbers, manual and heavy manual categories were combined in the analyses.

Leisure-time PA (recreational and household)

Detailed information was gathered on time spent (in hours/week) on several recreational activities, both in summer and winter: walking, cycling (either when going to work or as a recreational activity), gardening, do-it-yourself activities and sport (e.g., soccer, keep-fit, swimming, jogging, tennis, gym…). Household activity was also assessed as the total number of hours per week spent doing the housework (cleaning, doing the washing, cooking, taking care of children…), and the flights of stairs climbed per day were registered. A leisure-time physical activity variable was then computed weighing the time spent in each activity (walking, cycling, gardening, sport, do-it-yourself activities, housework and stair climbing) by its specific intensity score [23] and expressed as MET-h/week. For leisure-time activity, sex-specific quartiles were computed.

Vigorous PA

Participants were asked to declare the total amount of time (in hours/week) when the engaging in any of these activities had been vigorous enough as to cause sweating or a faster heartbeat. Vigorous PA was defined as none, ≤2 and >2 h/week.

Overall activity index

An overall PA index was computed (Cambridge/Bilthoven PA Index, CPAI) by combining occupational activity and time spent in sport and cycling, as previously described [24]. This index consists of four ordinal categories: inactive, moderately inactive, moderately active and active (see “Appendix”). Participants not working or with missing occupational data were ranked according to time spent in sports and cycling, only. The CPAI was validated against objectives measures of physical activity energy expenditure (PAEE) as determined by heart rate monitoring (Spearman’s rhoCPAI-PAEE = 0.27, p < 0.05; Ulf Ekelund, personal communication).

Follow-up procedures and endpoint assessment

Incident gastric and oesophageal tumours were identified through population cancer registries, except in France, Germany, Greece and Naples where other methods, including health insurance records, cancer and pathology hospital registries or active follow-up were used. Mortality data were based in national or regional mortality registries. Participants were censored at the date of diagnosis of a first incident tumour, death, loss to follow-up due to emigration or other cause, or end of follow-up, whichever came first. Closing date was that of the last complete follow-up, between December 2002 and 2005, depending on the study centre.

The 2nd edition of the International Classification of Diseases for Oncology [25] was used to code site (C15.0–C15.9: oesophagus, C16.0: gastric cardia and GEJ; C16.1–C16.8: stomach, non-cardia) and morphology (adenocarcinoma codes 8140/3, 8144/3 and 8145/3). The histological type was classified as intestinal or diffuse following the Laurén Classification [26].

An expert panel of pathologists was in charge of the validation of diagnosis and classification of tumours. The panel included a representative from each country participating in EPIC and a coordinator. Almost 60% of gastric tumours (n = 245) were histologically confirmed as adeoncarcinomas by the panel. Pathology reports and protocol forms were used for tumour classification in 23% of cases (n = 97), and 17% of gastric adenocarcinomas (n = 68) were classified according to the codes provided by the cancer registries to the IARC central database. For oesophageal adenocarcinomas, these figures were 69% (n = 55), 15% (n = 12) and 16% (n = 13), respectively. The methodological issues on the panel validation procedures and criteria have been detailed elsewhere [27].

Statistical analyses

Mean values (and standard deviations) or percentages were used as descriptive statistics (Kruskall–Wallis and χ2 tests were used to assess statistical significance). The risks of incident adenocarcinomas of the stomach and oesophagus were modelled through proportional hazards Cox regression, with attained age as the underlying time variable [28]. Entry time was defined as age at recruitment and exit time as age at case diagnosis or censoring. Models were stratified on age (in 1-year categories), and centre, to control for differences in study design, diagnosis and follow-up procedures. All analyses were adjusted for sex, height (m), weight (kg), educational level (no studies, primary, technical/professional school, secondary, university and missing), smoking status (never smoker, former smoker who quit ≥10 years ago, former smoker who quit less than 10 years ago, former smoker of unknown quitting, current smoker of <15 cigarettes/day, current smoker of 15–25 cigarettes/day, current smoker of ≥25 cigarettes/day, current smoker of unknown amount of cigarettes/day and missing), alcohol consumption at recruitment (g/day), energy intake (kcal/day) and daily consumption of fruit, red meat and processed meat (in g/day).

Separate regression models were defined to evaluate specific risks of gastric cancer by site (cardia and non-cardia) and histology (intestinal and diffuse). The interaction of the physical activity index with age and body mass index was evaluated with two-sided likelihood ratio tests comparing models with and without the interaction term. We explored the potential modifying effect of Helicobacter pylori infection on the association between PA and risk of GAC. For this purpose, we used data from a previous nested case–control study of H. pylori infection in regard to risk of gastric cancer carried out within the EPIC cohort [29]. Details on laboratory methods have been reported elsewhere [29]. Unconditional logistic regression was applied to derive odds ratios of incident stomach adenocarcinoma risks by PA separately by infection status in the subset of cases and non-cases not meeting the exclusion criteria.

Analyses of the physical activity index were repeated after exclusion of cases diagnosed within the first 2 years (n = 93) from the date of recruitment. Sensitivity analyses were also performed excluding subjects with self-reported diabetes or cardiovascular problems at baseline (n = 90,371).

The STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) guidelines were taken into consideration in the writing of the manuscript [30].

Results

For the present analysis, mean follow-up time was 8.8 years completing 3.7 million person-years. A total of 410 first primary incident GAC cases (85% of total incident gastric cancer cases) and 80 OAC (78% of the total) occurred during this time. The distribution of adenocarcinoma cases of the stomach and oesophagus by participating country is presented in Table 1. Distal adenocarcinomas of the stomach were more frequent than those of the cardia region. A similar number of intestinal and diffuse tumours were found, overall. In regard to OAC, almost all incident cases were diagnosed in northern European countries (Denmark, Sweden and United Kingdom), with very low number of cases in the Mediterranean area.

Table 1 Summary of incident adenocarcinomas of the stomach and oesophagus, by country
Full size table

Subjects in the highest quartile of recreational PA had the highest body mass index (BMI), the highest intake of fruit and vitamin C and the lowest consumption of alcohol, red meat and β-carotene (Table 2). These participants had also attained a lower educational level and were more likely to report diabetes or cardiovascular problems at baseline. Smoking status also differed by levels of PA. Most people in the upper PA quartile were never smokers, whereas the higher frequency of smokers of ≥15 cigarettes/day was found in the lowest PA quartile.

Table 2 Distribution of variables of interest according to sex-specific quartiles (Q 1–Q 4) of leisure-time physical activity
Full size table

The main results of the evaluation of the prospective association between PA and risk of incident gastric and oesophageal adenocarcinomas are shown in Table 3. Levels of occupational activity were not related to GAC or OAC. No association was found for occupational or overall leisure-time PA and risk of GAC (neither all types nor site- or histological-specific tumours). Overall PA as estimated by the CPAI showed a significant reduction in GAC risk between 25 and 38% for different categories of PA when compared to the inactive group (hazard ratio (95% CI) = 0.69 (0.50–0.94) for active vs. inactive participants). This lower risk was consistent across categories with a statistically significant linear trend (p = 0.006). This result was restricted to non-cardia adenocarcinomas of the stomach, whereas it was not observed for intestinal or diffuse types. Vigorous PA was not associated with cancer risk in this analysis. For any variable studied, no effect of PA on the risk of developing OAC was found. Nevertheless, a closer analysis of different recreational activities revealed that participants regularly engaging in cycling and sport were less prone to develop GAC, especially at distal locations, whereas the habit of walking was associated with an increased GAC risk (Table 4).

Table 3 Hazard ratio of gastric adenocarcinoma, by tumour site and histology, and overall oesophageal adenocarcinoma according to levels of physical activity in the EPIC cohort
Full size table
Table 4 Hazard ratio of gastric adenocarcinoma, by tumour site and morphology, and overall oesophageal adenocarcinoma according to levels of recreational physical variables in the EPIC cohort
Full size table

In our analyses involving n = 1,211 participants of a prior nested case–control study [22], Helicobacter pylori was detected in 83.2% of cases (n = 163) and 68.7% of non-cases (n = 697). In this subset of participants, a significant interaction was detected between H. pylori infection with type of occupation and recreational PA (Table 5). Nevertheless, there were no significant associations or a clear trend in regard to GAC in separate analyses for infected and not infected participants. The association between the total PA index and GAC persisted when adjusting for H. pylori seroprevalence.

Table 5 Odds ratio of incident gastric adenocarcinoma (GAC) by physical activity levels, according to Helicobacter pylori infection status
Full size table

Discussion

Overall physical activity (PA) was associated with a significant reduction in risk of developing total and non-cardia gastric adenocarcinomas in this study. No association overall was seen for increased PA in regard to OAC. Even when previous epidemiological data suggested the possibility of an independent association between physical activity and risk of tumours of the stomach and the oesophagus [11, 15, 16], the analysis of such relationship had not been sufficiently addressed in the context of large prospective studies until recently. Our study provides valuable results for gaining insight into this association [17–20].

Physical activity at work

A sizeable amount of the daily PA among workers is determined by the physical demand of their job. A sedentary job might add to a sedentary lifestyle in the promotion of overweight and obesity [6]. Nevertheless, we did not find a significant risk reduction in those subjects with manual or standing occupations when compared to sedentary workers, in regard to GAC or OAC. These null results agree with previous existing literature [15, 22] and do not support the finding by Severson et al. [14] relating gastric cancer risk with moderate or heavy working activities.

Leisure-time activity

The daily amount of recreational PA, probably the most variable component of energy expenditure, largely reflects individual preferences and health-based decisions (household activities are less subject to free choice). As such, recreational activity represents the key PA component in terms of prevention [31]. Whereas our study does not support an overall protective role for leisure-time PA (neither household nor recreational PA separately; data not shown), a more in-depth analysis suggested an inverse association for cycling and sporting activities (but direct for walking) and risk of GAC (Table 4). This association was site-specific, restricted to distal tumours, but unrelated to the histological type. It could be speculated that only physiologically demanding activities would be of a sufficient magnitude as to exert a relevant health effect [11, 32], but our data do not fully support this speculation. On the one hand, only those activities with high MET values were inversely associated with GAC (METcycling/sport = 6). On the other hand, the amount of vigorous PA was not associated with risk of GAC or OAC. This finding is in disagreement with previous evidence [16]. However, reported vigorous PA would not necessarily be expected to decrease the risk of the disease. An exercise would cause sweating or increased heartbeat when the intensity is high, but also when the physical fitness of the individual is poor, which would have a very different meaning in relation to cancer risk. Besides, high-intensity PA may be less prone to misclassification than other activities of light or moderate intensity (e.g., walking). Our results would support the practice of regular exercise of moderate to high intensity rather than episodes of exhausting activity.

Overall physical activity

The CPAI (see “Appendix”) was designed to capture relevant variability in overall PA by combining PA at work and recreational activities (sports and cycling), so that the ranking of participants reflected actual differences in PA [22]. Our results suggest a protective role of PA against GAC (but not OAC), showing a significant decrease in risk of non-cardia tumours by increasing PA categories of the index. The hazard ratio (HR) for the ‘active’ relative to the ‘inactive’ group was 0.44 (95% CI: 0.26–0.74; p for trend = 0.001). For total GAC, this HR was 0.69 (95% CI: 0.50–0.94). Few studies have analysed the effect of PA and risk of gastric cancer separately by site or histological type. A previous case–control study conducted in Canada estimated average lifetime PA of the participants and found this measure to be inversely associated with risk of total and cardia gastric cancer risk [11]. For non-cardia tumours, the age- and sex-adjusted inverse association found was weakened after controlling for a series of confounders in multivariate models. All those variables were controlled for in our analyses, except the Western dietary pattern. Nevertheless, we adjusted for those elements of this pattern of greater relevance for gastric or oesophageal cancer risk, such as consumption of red and processed meat, fruit, alcohol and total energy intake [6, 33, 34]. Other potential confounders tested (such as vegetable and fibre intake) had no significant effect on the associations under study (<5% coefficient change) and were finally discarded due to efficiency reasons. Therefore, it is not likely that the discrepant results could be attributed to residual confounding by diet. Only two previous prospective studies have reported results by site [18, 20]. In the first, recreational PA was associated with a significant reduction in non-cardia cancer risk [18]. Unfortunately, the authors did not include cardia tumours in their analysis. The second one showed an independent association for non-cardia adenocarcinomas only [20]. According to the available evidence, PA may be regarded as protective in relation to distal (non-cardia) forms of gastric cancer, although an effect would be more expected to be observed at cardia locations, which are probably associated with obesity [35]. Further prospective evidence might be needed before these results can be considered as conclusive.

The mechanisms by which increased physical activity could affect the risk of gastric or other types of cancer are largely unknown. Both generalised and site-specific effects have been alleged to account for a protection of PA on overall cancer risk [6, 31, 36–38]. An effect on chronic inflammation or the modulation of oxidative stress would be plausible mechanisms, but it is unclear whether PA is able to inhibit any (or both) of them. In our study, we did not find indications that the significant association of PA and stomach cancer risk could be mediated by an effect of PA on body fat deposition (adjustment or stratification by BMI categories did not reveal significant interactions). The same was reported by others for overall [18] and non-cardia GAC [20]. Although a potential influence of sex-related factors cannot be discarded, we found no significant interactions by sex overall (data not shown). Of interest is that, recently, increased physical activity has been associated with the methylation level of CACNA2D3, a gene involved in the regulation of cell cycle [39]. This gene was found to be hypermethylated (i.e., silenced) in cultured gastric cancer cell lines [40]. Whether PA would be capable of epigenetic silencing of CACNA2D3 would first need to be established.

We studied the potential modulation of the PA effect on gastric cancer due to infection by H. pylori. Chronic inflammation of the gastric mucosa is an early event in H. pylori-associated gastric tumorigenesis. Our results do not clearly support that infection by H. pylori would be able to modulate the effect of PA on GAC risk. Although interactions by infection status were significant for different types of PA, the low number of cases analysed does not allow us to draw firm conclusions on this topic. Since no other study has taken into consideration H. pylori seroprevalence when analysing PA and GC risk, further research is needed to elucidate this point.

The EPIC study provides a unique epidemiological framework for testing PA and cancer interactions. The size and geographical distribution of the cohort was aimed at maximising the variability in exposures and confounders, and the prospective design makes it possible to assess the long-term effect of PA. The CPAI has been validated against measurements of energy expenditure based in heart rate monitoring. Although the correlation was not impressive (r = 0.27), the index was consistent across centres and suitable for ranking individuals (Ulf Ekelund, personal communication). Another strength is that bias due to outcome misclassification has largely been eliminated because the majority of cases were confirmed by a panel of expert pathologists. Further considerations apply when interpreting these results. First of all, the limited statistical power due to the low number of endpoint cases, in spite of the long follow-up period. The assessment of PA was indirect, based on an administered questionnaire, and some degree of misclassification may exist. Medical advice to increase PA could have occurred in participants with chronic disease at baseline. We therefore performed sensitivity analyses excluding those participants reporting diabetes or cardiovascular problems at recruitment. Neither the exclusion of these patients nor the GAC cases diagnosed within the first 2 years of follow-up modified the results noticeably. In spite of reported differences in recreational PA [41] and prevalence of H. pylori infection [42] across European countries, we found no significant heterogeneity by region (North vs. South; data not shown). The use of sex-specific cut-offs and the stratification by study centre should account for the variability in leisure-time PA by sex and country. A concern in relation to age arises when considering that ‘inactive’ participants were much older, on average, than those classified in any other PA category. While not an unexpected finding (older people reporting less PA), it is a disturbing one if taking into account that the ‘inactive’ group was set as the reference category for analyses. We cannot discard here the possibility of reverse causation if older people suffering from chronic gastritis (precursor of GAC) would have reduced their level of PA. In order to control for its effect in Cox models, we defined age as the underlying time variable and further stratified the models by 1-year categories, allowing comparisons within age strata instead of simply controlling for its effect in a cohort-wide model. In stratified analyses, the association with the CPAI remained significant only in the 55- to 60-year-old group, although there was no heterogeneity overall (data not shown). Finally, as in any observational study, the possibility of residual confounding cannot be totally discarded.

In conclusion, we present prospective data showing a decreased risk of gastric adenocarcinoma associated with higher levels of PA. This association manifested principally in non-cardia tumours, whereas no effect was found for PA in relation to intestinal/diffuse forms of GAC or OAC. Further studies should be aimed at clarifying whether these associations would be causal.