We found that users of NSAIDs, in particular those reporting current use, experienced a statistically significant lower risk of EAC than those who did not use these medications. We also observed statistically significant inverse associations between greater frequency and duration of NSAID use and risk of EAC. When the analysis was examined by type of NSAID, the magnitude of the effect for ever use of aspirin was similar to the effect for ever use of non-aspirin NSAIDs. However, there was little evidence of a dose-response effect when examining type of NSAID by increasing frequency or duration. The patterns and overall inverse association between NSAID and aspirin use and risk of EGJA appear to be similar to EAC but less striking.
An inverse relationship between NSAID use and esophageal and gastric adenocarcinomas has been reported previously in reviews and meta-analyses 27-31
. An important advantage of our pooled study is that we were able to create more standardized categories of exposure using individual-level data from each study and harmonized analytic models with respect to covariates for adjustment of potential confounding; these steps permit more comparable data than is possible using only results from published odds ratios, which utilize different sets of covariates between studies. We also built on past reviews by evaluating additional aspects of NSAID use, such as frequency, duration, and temporal factors, in a larger number of cases and controls than previous reports. In the most recently published meta-analysis, Abnet et al. reported that any aspirin use was inversely associated with EAC and EGJA cancers with summary ORs of 0.64 (95% CI=0.52-0.79) and 0.82 (95% CI=0.65-1.04), respectively 29
. Both are similar to the estimates we observed from our pooled study. We consistently observed that the effect of aspirin and NSAID use was slightly weaker for EGJA than EAC. Although only a few reviews evaluated both EAC and EGJA 29, 30
, the summary estimates reported are consistent with our observations of a stronger association with EAC. This is also consistent with past meta-analyses that reported a statistically significant reduced risk for gastric non-cardia cancer with NSAID use but a weaker effect for EGJA 28, 31
. Since these meta-analyses were published, one additional prospective study found an inverse association between regular use of aspirin and risk of EGJA 26
. Four additional studies conducted among high-risk cohorts of patients with Barrett's esophagus, two prospective and two retrospective, have also observed evidence of a reduced risk of progression to EAC with NSAID use 19, 38-40
. A pooled analysis of three randomized trials of daily aspirin use and at least 5 years of treatment found a delayed but significant reduction in deaths due to EAC and EGJA after 5+ and 10+ years of post-trial follow-up, respectively 41
. Although, a small, short-term trial of celecoxib/placebo among subjects with Barrett's esophagus found no change in the proportion of dysplastic biopsies between the treatment (200 mg/pill, 2 pills/day) and placebo group at 48 weeks42
, almost half of these patients already had dysplasia at study entry; a much larger randomized trial, including adequate numbers of patients without dysplasia, is needed in order to fully explore this hypothesis. A randomized trial of aspirin and proton pump inhibitors among patients with Barrett's esophagus is currently ongoing in the United Kingdom 43
. Results from this trial will assist in the ongoing effort to further examine the relationship between aspirin and the development of EAC.
In our study, the association with ever use of aspirin and non-aspirin NSAIDs was generally similar for EAC risk. This is in agreement with findings from other reviews that evaluated the effect of NSAIDs by type. All of our analyses were mutually adjusted for the other type of NSAID, which did not change any of the observed associations substantially. This suggests that although similar in effect estimates, both aspirin and non-aspirin NSAID use appear to have an individual effect on EAC risk above and beyond use of the other type of NSAID. That is, exposure to any member of this class of medication further reduces the risk of EAC. The similar effects of aspirin and non-aspirin NSAID use is substantiated further by the nearly identical associations observed between individuals who reported using aspirin only and individuals who reported using non-aspirin NSAIDs only with EAC and EGJA risk. As experimental studies have reported similar pathways and actions of aspirin and other NSAIDs, such as inhibiting COX-2, decreasing inflammation, increasing apoptosis and decreasing proliferation, it seems reasonable that there would be little difference between their effects on EAC or EGJA risk 44, 45
Frequency of NSAID use appeared to be more strongly associated with a reduction in EAC risk than duration of NSAID use in this analysis. Our analysis found an inverse relationship with increased frequency of NSAID use (p-trend<0.001), but a consistent reduced risk of EAC with any level of duration. A similarly strong inverse dose-response between frequent NSAID use and esophageal cancer was also reported in the meta-analysis by Corley et al. 27
. None of the previous EAC meta-analyses evaluated the effect of duration of use so we were unable to compare our results with others. Although we observed an association with risk of EAC, we found no significant associations among the highest categories of frequent or longer duration of NSAID use and risk of EGJA. In a meta-analysis by Wang et al., an inverse association with “regular” or frequent use of NSAIDs and gastric cancer was observed; however, there was inadequate evidence to suggest an association with duration of NSAID use in their analysis 28
. The interpretation of results on frequent NSAID use from the Wang et al. meta-analysis, however, is limited due to the inclusion of a mixture of gastric cancer sites and varying definitions of “regular” NSAID use across studies included in their meta-analysis.
Although analyses were limited to a subset of studies, in our study it appears that the reduced risk of EAC and EGJA was restricted to those who reported current use of NSAIDs. Past reviews have not attempted to summarize the effect of current NSAID status on risk of either EAC or EGJA. This is most likely due to the lack of published estimates, whereas we were able to utilize raw data from three of the six BEACON studies to construct this variable. The current NSAID users in our study tended to report frequent NSAID use (median = ~1 pill/day) and a moderate duration of use (median = 5 years). A similar pattern with current NSAID status has been observed in the one prospective study of Barrett's esophagus patients. Vaughan et al.19
observed that current NSAID use at baseline was associated with a significantly lower risk of EAC, which was stronger in a subset of patients who were deemed to be current users for the full duration of follow-up (median = 5.5 years).
Results from our study could suggest a possible short-term protective effect of NSAIDs on EAC and EGJA risk as we observed an effect with current users, higher frequency but not longer duration. However, as our pooled analysis consisted primarily of data from case-control studies, some of the observed effect could be due to reverse causation. Continued NSAID use can lead to erosive tissue damage in the gastrointestinal tract, which may result in patients with gastroesophageal reflux discontinuing NSAID use. However, the study's supplemental analyses did not support this possibility; a similar effect of aspirin and NSAIDs use and risk of EAC and EGJA was observed among individuals with or without gastroesophageal reflux. In addition, findings from the NIH-AARP study, a prospective cohort included in this analysis, were similar to findings of EGJA risk among case-control studies in this analysis. If replicated in further prospective studies and randomized trials, this could have implications on the use of NSAIDs as a potential preventative measure.
Our pooled study has a number of strengths as well as limitations. It included and combined data from six population-based case-control and cohort studies. This allowed for a large sample size and suitable statistical power to evaluate overall main effect associations, though the case numbers were still small in some strata and limited our power to fully evaluate effect modification. Combining data from a subset of our studies with extensive questions on usage also gave us the ability to evaluate characteristics of NSAID use which has been limited in past reviews. There was some variability among questions from different studies; in particular, we noticed two general patterns in how questions were asked. When the lead-in question included the definition of “regular use,” the prevalence of NSAID use among controls was around 40%; while broader questions reported a prevalence of around 70% among controls (Appendix A
). Despite the two patterns of questions, results stratified by question type demonstrated consistently inverse associations. There were also some minor differences on how frequency of use was collected: open-ended vs. pre-determined responses, and what time period served as the reference period for estimating usage (Appendix A
). Assuming the time period was representative of usual usage, we were able to define NSAID use in the same way across studies. Detailed individual-level data on known and suspected risk factors for EAC and EGJA were available for pooling from each study, thus allowing us to control for possible confounding from a standard set of risk factors across each study. There was some evidence of moderate heterogeneity between study populations; however, overall, there appears to be little evidence of heterogeneity across the majority of results except for duration of aspirin. Overall NSAID duration might have been overestimated, since we were unable to take into account potential concurrent use of aspirin and non-aspirin NSAIDs in the calculation. However, we re-evaluated overall NSAID duration taking into account concurrent use in two studies that collected sufficient data to carry out this analysis, and found that there was very little difference in study-specific estimates. Another limitation of our data is that no dosage information was collected in any of the studies. Inclusion of NSAID amounts would have provided a better indicator of drug exposure than frequency and duration alone. As most participating studies were case-control studies, data from these studies may be subject to biases; however, data from each study were obtained rapidly after enrollment and collected with detailed procedures. Results were also generally similar to those of the cohort study. Finally, we were unable to directly evaluate whether NSAID use varied by Barrett's esophagus status and whether this would have had an effect on reported associations, as the number of cases with information on Barrett's esophagus diagnosis (n=93) was not large enough to conduct any suitably powered analyses. However, with the small proportion of cancer cases with a prior Barrett's diagnosis, any effect on the overall results is likely to be small.
In summary, findings from this pooled analysis support the hypothesis that NSAID use provides potential benefits in preventing esophageal and esophagogastric adenocarcinomas. There is also evidence to suggest that increased frequency and longer duration of overall NSAID use is associated with further reduced risk of EAC, but these associations are not present when evaluated by NSAID type. Results from stratified analyses indicate that aspirin and non-aspirin NSAIDS have a similar effect on EAC risk overall. While the use of NSAIDs as a chemopreventive strategy is promising, the effectiveness of NSAIDs still has to be evaluated in randomized trials before this approach is advanced clinically.