Article Text
Abstract
Objective: To measure the extent to which risks of oesophageal cancers associated with gastro-oesophageal reflux (GOR) are modified by common factors including smoking, non-steroidal anti-inflammatory drugs (NSAIDs) and acid suppressant medications.
Design and setting: Population-based case–control study.
Participants: Cases were patients with oesophageal (OAC; n = 365) or gastro-oesophageal junction (GOJAC; n = 426) adenocarcinomas, or squamous cell carcinomas (OSCC; n = 303). Controls were sampled from a population register (n = 1580).
Main outcome measure: Odds ratio and 95% confidence interval.
Results: Frequent (at least weekly) symptoms of GOR were associated with significant 6.4-fold, 4.6-fold and 2.2-fold increased risks of OAC, GOJAC and OSCC, respectively. Under models examining effects of combined exposure, patients with frequent GOR symptoms who were also heavy smokers had markedly higher OAC risks (OR = 12.3, 95% CI 6.3 to 24.0) than those with frequent GOR who did not smoke (OR = 6.8, 95% CI 3.6 to 12.9). Similar patterns were observed for GOJAC and OSCC. Among people with frequent GOR symptoms, regular use of aspirin/NSAIDs was associated with almost two-thirds lower OAC risks (OR = 4.8, 95% CI 2.5 to 9.2) than non-users (13.9, 95% CI 6.5 to 30.0). In contrast, among those with frequent GOR symptoms, users of acid suppressants had similar OAC risks (OR 7.8, 95% CI 5.2 to 11.8) to non-users (OR 5.3, 95% CI 3.2 to 9.0).
Conclusions: People experiencing frequent GOR symptoms have markedly increased risks of OAC and GOJAC, and this effect may be greater amongst smokers. Use of aspirin and NSAIDs, but not acid suppressants, significantly reduced the risks of oesophageal cancers associated with GOR.
Statistics from Altmetric.com
The incidence of adenocarcinomas of the oesophagus (OAC) and the gastro-oesophageal junction (GOJAC) has been rising rapidly in Western populations to the extent that adenocarcinomas are now more common than oesophageal squamous cell carcinoma (OSCC) in many nations.1 2 3 Change over time in the prevalence of exposure to environmental factors is the most likely explanation for the observed trends. Epidemiological studies consistently identify people with frequent symptoms of gastro-oesophageal reflux (GOR) as having the highest risks of OAC, with most studies reporting significant dose–response effects.4 5 6 Recently, we noted a synergistic interaction between GOR and obesity which led to markedly increased risks of OAC among people having both conditions4 7 and it is possible that such effects have contributed to the rising incidence of OAC.8 9
Other factors have been identified which appear to strongly determine the risk of oesophageal cancer, and it is possible that such factors may modify the risks associated with of GOR.10 Smoking is a risk factor for all oesophageal cancers,11 12 13 and while an interaction between GOR and smoking has been observed for Barrett’s oesophagus (a precursor condition for OAC),14 this has not been explored for OAC. People who regularly consume aspirin and non-steroidal anti-inflammatory drugs (NSAIDs) have been observed to have significantly lower risks of OAC, GOJAC and OSCC than people who do not take these medications.15 16 17 Inhibition of COX-2 and related inflammatory pathways has been suggested as the mechanism to explain this apparently protective effect of aspirin and NSAIDs; if so, then this effect should be particularly apparent among people with inflammation of the lower oesophagus as characterised by frequent GOR. Patients with frequent GOR symptoms are also commonly treated with medications to suppress the production of gastric acid, including H2 receptor blockers and proton pump inhibitors (PPIs). If repeated exposure of oesophageal epithelium to gastric acid is the underlying cause of OAC, then it might be predicted that among patients with frequent GOR, those taking acid suppressant medications would have lower risks of OAC than those not taking such medications. To date, there are scant data with which to draw conclusions on these issues.
Here, we report the findings of an investigation into the association between GOR symptoms and oesophageal cancer, and which specifically sought to identify factors that modify the association.
Patients and methods
Study design and participants
We analysed data from a nationwide case–control study of oesophageal cancer conducted in Australia which was approved by the research ethics committees of the Queensland Institute of Medical Research (QIMR) and participating hospitals. Details of recruitment were published in an earlier report.4 In summary, eligible case patients were people aged 18–79 years with a histologically confirmed primary invasive cancer of the oesophagus or gastro-oesophageal junction diagnosed between 1 July 2002 (1 July 2001 in Queensland) and 30 June 2005 in the mainland states of Australia. Patients were recruited either through major treatment centres or state-based cancer registries. A total of 1577 patients with oesophageal cancer received an invitation to participate in the study, of whom 1102 patients returned a completed questionnaire (70% of all invited; 35% of all living and deceased persons in mainland Australia who had been diagnosed with incident oesophageal cancer). For these analyses, the final numbers of case participants were 365 OAC, 426 GOJAC and 303 OSCC patients. Controls were randomly selected from the Australian Electoral Roll (enrolment is compulsory) and matched to the cases within strata of age (in 5 year age groups) and state of residence. Of 3258 potentially eligible control participants who were contacted and invited to participate, 175 were excluded (16 died, 61 too ill to participate, 98 unable to read or write in English), and 41 were lost to follow-up after initial contact. Of 3042 remaining controls, 1680 (55%) accepted the invitation and 1580 (51% of all potentially eligible controls contacted) returned completed questionnaires. Complete data for all variables included in the statistical models were available for 1545 controls, 344 OAC, 405 GOJAC and 277 OSCC.
Data collection: demographic information and health survey
Participants self-completed a health and lifestyle questionnaire, which asked about their education, occupation, general health, height and weight at the reference age (1 year ago for controls and 1 year before diagnosis for cases), smoking history, alcohol, use of aspirin and NSAIDs. We elicited a history of GOR symptoms by asking about experience of heartburn (“a burning pain behind the breastbone after eating”) or acid reflux (“a sour taste from acid or bile rising up into the mouth or throat”). Positive responses were followed by questions asking the age of first experiencing symptoms and the frequency of symptoms at ages 10–19 years, 20–29 years, 30–49 years and 50–79 years in five categories. Participants were also asked if they had ever used, separately, aspirin, other NSAIDs or paracetamol during the past 5 years, and if so, the frequency of use on a seven-point scale ranging from less than once a month up to two or more times/per day. Checklists of generic and trade names of medications licensed for use in Australia were provided to aid recall. Although aspirin and NSAIDs use were asked separately, our prior analyses have shown similar effects of the two, hence they were combined for these analyses.18 We also asked participants to indicate whether they had ever consumed any of eight classes of medications (generic name and all brand names licensed for use in Australia at the time) used to treat heartburn, acid reflux or dyspepsia. For these analyses, we focused on participant responses to the two classes of acid suppressant medications (specifically, H2 receptor blockers and proton pump inhibitors (PPIs)). We combined the responses into a single binary variable (“ever/never use of acid suppressant medication”) after preliminary analyses demonstrated no differences in the magnitude of effect.
Derivation of variables for analysis
Participants who reported no symptoms of either heartburn or acid reflux during any age period were defined as never having GOR symptoms; all other participants were assigned a GOR symptom frequency equal to the frequency of either heartburn or acid reflux, whichever was the highest. We collapsed the frequency of symptoms into three categories (“never”, “less than weekly” and “at least weekly”) for analysis. Duration of GOR symptoms was calculated as the difference between the reference age and the lower boundary of the age interval in which regular (ie, at least weekly) GOR symptoms were reported for the first time. Body mass index (BMI) was calculated by dividing the weight, in kilograms, by the square of height, in metres. We obtained the lifetime smoking dose in pack-years by summing the smoking dose over each decade of life, calculated by multiplying the decade specific smoking intensity during that decade (cigarettes/day) by the number of days of smoked per week and the smoking duration within that decade.
Statistical analysis
Our primary aim was to measure the relative risk of oesophageal cancer associated with GOR symptoms, and then to assess the effect of potential modifiers. We fitted multivariable logistic regression models to calculate the odds ratios (OR) and the 95% confidence intervals (95% CIs) for the association between GOR symptoms and the three cancer outcomes. We first analysed the frequency of GOR symptoms at four different age intervals (10–19, 20–29, 30–49 and 50–79 years) to examine whether the risk estimates varied with symptoms at different stages of life. Thereafter, we focused principally on the frequency of GOR symptoms reported by each participant during the questionnaire age interval coinciding with 10 years before the reference age. This historic period of exposure was selected to minimise erroneous reporting of cancer symptoms among cases and because of the likely long latency of effect of acid exposure.
Base models included terms only for age and sex, to which we added terms for education, smoking, use of aspirin/NSAIDs in the past 5 years, BMI and alcohol consumption. The matching variable, state of residence, was included in early models but made no difference to any results and hence was dropped. We examined potential statistical interactions between GOR symptoms and smoking, aspirin/NSAIDs use in the past 5 years and other medications by including relevant multiplicative terms in the multivariate model. We also derived measures of combined exposure by cross-classifying each participant according to their GOR symptom frequency and their level of exposure to the potential modifiers (viz. smoking, aspirin/NSAIDs, and acid suppressant medications). We estimated the relative risks associated with exposure to GOR symptoms alone, potential modifiers alone, and both. From these risk estimates, we calculated the synergy index (S), which compares the observed excess risk for the combined exposure category relative to the expected excess risk assuming that GOR and the potential modifiers are independent risk factors under an additive model.19 All analyses were conducted in SAS (V9) and all significance tests were two sided test at α = 0.05.
Results
Demographic and GOR distribution
The distribution of demographic characteristics among cases and controls are presented in table 1. Cases were less likely than controls to be university educated and more likely to be ever smokers. Higher proportions of OAC and GOJAC cases were obese (BMI ⩾30) compared to controls, whereas the opposite was observed for OSCC cases. Controls were slightly more likely to report ever having used aspirin/NSAIDs than cases. Among controls, use of aspirin/NSAIDs was more common among those reporting frequent reflux symptoms compared to those reporting no reflux symptoms. Use of acid suppressant medications (H2 blockers, PPIs) was reported by substantially higher proportions of cases than controls.
Sixty-seven per cent of controls reported ever having experienced GOR symptoms compared to 84% of OAC and 79% of GOJAC cases. The proportion ever having GOR symptoms was lower among OSCC patients. The frequency of GOR symptoms generally increased with each successive age category for each of the case and control groups (table 2). OAC cases were more likely to report at least weekly GOR symptoms at all ages.
Overall risk estimates associated with GOR symptoms
We observed significant 2- to 3-fold increases in the risk of OAC (OR, 2.7; 95% CI, 2.0 to 3.8) and GOJAC (OR, 2.1; 95% CI, 1.6 to 2.8) among those who had ever experienced GOR symptoms, but not for OSCC (OR, 0.9; 95% CI, 0.7 to 1.2). While the patterns of risk associated with GOR symptoms at different age intervals were consistent for all three cancers, the magnitudes of the relative risks of OAC were higher for those having GOR symptoms at earlier ages (table 2). People reporting at least weekly GOR symptoms after age 20 years had 6- to 7-fold elevations in risk for OAC. Those reporting at least weekly GOR symptoms before age 20 had approximately 10-fold increased risks of this cancer (OR, 10.5; 95% CI, 5.1 to 21.6). The overall risks of OAC and GOJAC among people with at least weekly GOR symptoms in the age category 10 years prior to diagnosis were increased by more than 6-fold (OR, 6.4; 95% CI, 4.6 to 9.1) and 4-fold (OR, 4.6; 95% CI, 3.3 to 6.3) respectively (table 2). In contrast, modest 2-fold elevations in risk of OSCC were observed for people reporting frequent GOR symptoms at all ages, including in the age category the past 10 years.
Combined effects of GOR symptoms and smoking
The risk estimates for GOR were not materially affected by adjusting for smoking, and vice versa, nor were the multiplicative interaction terms statistically significant. However, when interaction was assessed by comparing relative risks for smoking alone, GOR alone and both smoking and GOR, we observed departures from simple additivity. Thus those who reported a 30+ pack-years smoking history but no GOR symptoms experienced relative risks of 2.5 for OAC, 2.7 for GOJAC and 3.6 for OSCC (table 3). Among never smokers, those who reported at least weekly symptoms of GOR had markedly elevated relative risks of OAC and GOJAC (table 3). Relative risks due to the combined effects of GOR and smoking were 60% higher than expected under simple additive models for OAC (S, 1.6; 95% CI, 0.8 to 3.0), 150% higher for GOJAC (S, 2.5; 95% CI, 1.3 to 4.9) and 140% higher for OSCC (S, 2.4; 95% CI, 1.1 to 5.3).
Combined effect of GOR symptoms and aspirin/NSAIDs use
Regular use of aspirin/NSAIDs in the 5 years prior to diagnosis was associated with significant risk reductions of around 40% for all three cancers (table 4). Multiplicative interaction terms for aspirin/NSAID use and GOR were not statistically significant for any type of oesophageal cancer (p values for interaction were 0.21, 0.12, 0.44 for OAC, GOJAC and OSCC, respectively). However, in analyses of combined exposure to GOR symptoms and aspirin/NSAIDs (table 4), it can be seen that the risks of OAC and GOJAC were positively associated with frequency of GOR symptoms within all three categories of aspirin/NSAIDs intake (never; <1/week; ⩾1/week). However, the risks of OAC and GOJAC associated with at least weekly GOR symptoms were markedly lower among the regular aspirin/NSAIDs users compared to never users (OAC: OR, 4.8 vs 13.9; GOJAC: OR, 3.1 vs 8.5, table 4). These relative risks for combined exposure were 70% lower than expected under the additive model (S, 0.3; 95% CI, 0.1 to 0.7 for both OAC and GOJAC). While the risk of OSCC associated with frequent GOR symptoms diminished to null among regular users of aspirin/NSAIDs, there was no evidence of synergy. When the analysis for aspirin and NSAIDs were performed independently and exclusively, there were some fluctuations in the estimates of risks due to smaller sample size however the pattern of effects remained the same for both aspirin and NSAID.
Combined effect of GOR symptoms and acid suppressants
Use of acid suppressant medications was significantly more common among all groups of cases than population controls (see overall effects, table 5). As expected, these medications were used predominantly, although not exclusively, by people reporting GOR symptoms. We found no evidence that the risks of oesophageal cancer associated with GOR symptoms were modified by use of acid suppressants. Indeed, among those reporting at least weekly symptoms of reflux, the risks of OAC, GOJAC and OSCC were very similar for acid suppressant users and non-users (OAC: OR, 7.8 vs 5.0; GOJAC: OR, 4.9 vs 5.0; OSCC: OR, 2.7 vs 2.3) and the synergy index for the combined effect were not statistically significant. As our results may have been confounded by medication use due to symptoms shortly before diagnosis, we repeated our analysis in the subset of people where medication use in the past one to five years was available and reached the same conclusion (data not shown).
Discussion
We have explored the effect of GOR symptoms on the risks of the three most common subtypes of oesophageal cancer. We found consistently increasing risks of OAC and GOJAC with increasing frequency of GOR symptoms. Whilst no such trend was observed for OSCC, we did find an increased risk of OSCC among those who reported at least weekly frequent GOR symptoms, albeit of markedly lesser magnitude than those observed for OAC and GOJAC. Most importantly, we found evidence that smoking and use of aspirin/NSAIDs modified the effects of acid reflux in different directions. Specifically, whereas risks of OAC and GOJAC associated with GOR symptoms were higher among smokers compared to never smokers, they were significantly lower among regular users of aspirin or NSAIDs. We found no evidence that regular use of acid-suppressants modified the risk of OAC and GOJAC associated with frequent GOR symptoms.
Our finding of greater than 2-fold increase in the risk for OAC among those reporting ever having GOR symptoms is consistent with the reports from previous case control studies.5 6 Similarly, the 7-fold risk of OAC we observed among those reporting at least weekly GOR symptoms also agrees with previously reported risk estimates.6 20 21 Few studies have reported on associations between GOR symptoms and OSCC, and while we found a modest positive association, this was not observed in previous population based case–control studies conducted in the USA and Sweden.6 20
Smoking has long been known to be associated with increased frequency of GOR symptoms22 and the independent effects of these exposures on OAC have been previously documented;5 20 23 24 however, little was known about the effects of combined exposure. Our data suggest that people with frequent GOR symptoms who also smoke heavily have more than 1.5 times higher risks of OAC and more than two times higher risks of GOJAC and OSCC than would be predicted assuming that the risks of reflux and smoking were simply additive. In previous analyses, we found smoking to be more strongly associated with GOJAC than OAC13 and here we observed that the effects of combined exposure to smoking and GOR symptoms were also stronger for GOJAC than OAC. Further, we have previously reported the synergistic effects of gastro-oesophageal reflux and obesity on risks of OAC and GOJAC.4 While we adjusted for body mass index in all models, we cannot exclude the possibility that the complex patterns of synergy that we have identified in this set of analyses may differ according to adiposity. To identify and measure such multi-level interactions with precision would require a substantially larger sample, as might be achieved with a pooled analysis of existing datasets.
The biological mechanisms underpinning the association between gastro-oesophageal reflux and oesophageal cancer are becoming clearer. Chronic reflux of acid and bile into the oesophagus injures the epithelium, inducing cascades of cytokine responses25 and in turn leading to inflammation and cell proliferation. Experiments conducted in cell lines have demonstrated the genotoxic effects of exposure to physiological levels of acid and bile, including single- and double-strand DNA breaks and oxidative damage.26 27 28 In addition, acid in the oesophageal lumen reacts with nitrites in swallowed saliva to generate nitric oxide,29 a mutagen which has been demonstrated specifically to induce DNA damage in Barrett’s oesophagus cell lines.26 30
One of the mechanisms by which smoking might enhance the effects of GOR is through inflammatory pathways.31 Smoking is also known to relax lower oesophageal sphincter tone, leading to prolonged acid exposure32 which may enhance the effects of GOR and thereby increasing the risk for oesophageal cancer.
The likely chemopreventive role of NSAIDs has been described for many epithelial cancers,33 34 including oesophageal adenocarcinoma and its precursor Barrett’s oesophagus.15 16 17 35 36 Our estimates of approximate 40% risk reductions for all oesophageal cancers among frequent users of aspirin/NSAIDs were similar to those estimated by meta-analyses.16 36 Several previous studies have investigated possible interactions between use of NSAIDs and history of oesophageal disorders (including GOR), with some inconsistency of findings. Linblad et al37 observed greater risk reductions associated with NSAID use among those with upper gastrointestinal disorders, as we did. In contrast, Farrow et al15 and Anderson et al17 reported greater risk reductions among those with no reflux symptoms. Aspirin and NSAIDs reduce the severity of inflammation associated with GOR, which may protect the epithelium from possible carcinogenic sequelae. This phenomenon has been observed in studies of animals with surgically induced reflux, where COX-2 inhibitors reduced the degree of inflammation, progression to Barrett’s oesophagus, and eventual adenocarcinoma.34 Clinical trials in humans are currently under way to test whether aspirin prevents the development of OAC among patient with Barrett’s oesophagus, but the results are not yet available.38 Our data suggest that the possible benefits of aspirin should be greatest in those with the most frequent symptoms of reflux.
In contrast to the reduced risks of OAC and GOJAC associated with use of aspirin or NSAIDs, we found no evidence that use of acid suppressant medications such as PPI or H2 blockers modified the association between frequent GOR symptoms and OAC or GOJAC. While misclassification of medication history is always possible, our findings of specific associations with one class of medications (NSAIDs) but not others (acid suppressants) would argue against universally biased reporting of medication history. Indeed, for bias to explain our findings, the patterns of association we observed would require systematic under-reporting by cases of their NSAIDs exposures, but not H2 or PPI exposure, a pattern of recall we consider unlikely. We cannot exclude the possibility of confounding by indication, whereby patients with frequent reflux symptoms either were not prescribed or otherwise avoided the use of NSAIDs in the years preceding the study, although why this should apply only to cases but not controls is not easy to explain. Use of H2 blockers alone was shown to have no impact on OAC risk in one previous US study20 and we observed similar effects when we repeated our analyses restricted to those who used H2 blockers but not PPIs. While PPI use has been shown to be beneficial with regard to cancer risk in some studies,39 40 a clinical trial assessing short-term PPI treatment in patients with erosive oesophagitis observed changes in some measures of cellular immunity but found no difference in oxidative DNA damage, suggesting that reflux-induced genotoxicity is not altered by these medications.41 Our observation of overall higher PPI use among cases than controls may reflect confounding by indication, where PPI use is an indicator for symptom severity. Overcoming this potential bias is difficult using the case–control design; hence caution is required in interpreting such findings.
Strengths of the study include the large sample size, the population-based sampling frame and the very detailed measures of exposure to a broad range of potential causal factors. These data allowed us to explore modifiers of the association between GOR symptoms and oesophageal cancer to an extent not undertaken previously.
Our study experienced a low participation rate among controls, similar to other recent studies; however, the prevalence of at least weekly GOR symptoms in our control population (∼12%) was similar to that observed in other population-based studies in Australia,42 the UK43 and Sweden.6 Although recall bias may have occurred, several features suggest that the likely extent is limited. We obtained a detailed lifetime history of reflux, asking for the frequency and severity of symptoms within four discrete age periods. We found that very few participants reported GOR symptoms in early life, and that frequency of GOR symptoms increased with age among cases and controls, as reported in other studies.8 Moreover, we observed very different risk estimates for OAC and OSCC associated with GOR symptoms, strong evidence against systematic over-reporting among all patients with oesophageal cancer. Our finding of modestly elevated risks of OSCC associated with reflux has not been reported previously, however, and might be interpreted by some as evidence for biased recall. Although we cannot completely exclude this possibility, the patterns of association and subsequent effect modification by smoking and use of NSAIDs, point to an alternative conclusion. Another explanation for the associations with GOR might be “reverse causality”. This would occur if symptoms of oesophageal discomfort due to incipient cancer were misreported by cases as GOR. While this might partially explain the higher prevalence of recent GOR symptoms among cases, it is highly unlikely that symptoms from a growing tumour would be experienced at young ages, as we observed among OAC, GOJAC and OSCC cases.
Finally, we identified largely concordant findings in the relations between reflux, smoking and the use of aspirin/NSAIDs and risks of OAC and GOJAC. The overall patterns of association were very similar for these two cancers, although the associations with reflux symptoms appeared consistently stronger for patients with OAC than GOJAC, whereas associations with smoking were stronger for patients with GOJAC than OAC. It is possible that some of these differences in effect may stem from anatomical misclassification of tumours, since there is always uncertainty as to the precise location of origin of bulky tumours in the region of gastro-oesophageal junction.44 Thus a small proportion of the tumours defined as “GOJ” in this series may have originated in the proximal stomach, which might account for some of the differences in effect that we observed.
In summary, we have demonstrated the importance of GOR as a major risk factor for OAC and GOJAC, and quantified the strength of the association. Further, our data suggest that GOR symptoms might also be associated with OSCC, an association not identified in previous studies. Finally, we have found that the association between GOR symptoms and oesophageal cancer is exacerbated by smoking, and ameliorated by anti-inflammatory medications. Definitive evidence for these effects can only be obtained from experimental designs involving animals (to test possible synergies between smoking and reflux) or humans (to assess the effects of aspirin/NSAIDs among patients with frequent, chronic reflux).
REFERENCES
Footnotes
Funding This study was supported by the Queensland Cancer Fund and the National Health and Medical Research Council (NHMRC) of Australia (Program no. 199600). DW and PW are supported by Research Fellowships from the NHMRC. NP and SS were supported by PhD scholarships from the NHMRC and the Ministry of Health and Medical Education of the Islamic Republic of Iran, respectively. The funding bodies played no role in the design or conduct of the study; the collection, management, analysis, or interpretation of the data; or preparation, review, or approval of the manuscript.
Competing interests None.
Ethics approval This study was approved by the Human Research Ethics Committee of the Queensland Institute of Medical Research (Project-ID P399) and all participants gave their written consent to participate.
Authors’ contributions: NP performed the statistical analysis and prepared the manuscript. SS assisted in data preparation and interpretation. DW, PW and AG designed the original study and supervised the collection of data. All authors assisted in preparing the manuscript and approved the final version.
Provenance and Peer review Not commissioned; externally peer reviewed.