Should Patients on Long-term Low-Dose Aspirin Receive PPIs?
Should Patients on Long-term Low-Dose Aspirin Receive PPIs?
The current meta-analysis showed that concomitant PPI use in patients receiving low-dose aspirin confers a 73% reduced risk of peptic ulcer. However, the clinical significance of this effect is questionable. Peptic ulcer in all seven included studies were detected by endoscopy, but the associated symptoms were not reported in five of these studies. One study was conducted in asymptomatic patients and another reported that only 47% of patients had abdominal symptoms.
The pooled effect of PPIs on peptic ulcer should be interpreted with caution. The presence of significant heterogeneity indicates that this estimate may be biased. There was methodological diversity as we included both RCTs and cohort studies. Quality assessment of the studies revealed clinical heterogeneity in the treatments and baseline characteristics of the included patients. For example, the inclusion criterion on the minimal exposure time of the patients to aspirin and PPIs differed considerably between the observational studies, ranging from 7 days to 1 year for aspirin use and from 7 days to 3 months for PPI use. Types and doses of PPIs also differed between studies. One study selected only patients older than 65 years while the others included patients with different age ranges (min age, 18; max age, 92) and averages (67.4–79.7). The patient profiles of CV risk factors also varied across studies, partly because three studies did not exclude patients at very high CV risk while the others did. Age and the presence of CV risk factors may be confounders as they are likely to be associated with increased risks of GI events as well as with PPI use. Other potential confounders include a history of peptic ulcer disease and lower socioeconomic status; the latter is associated with Helicobacter pylori infection and the risk of peptic ulcer disease, and also with less access to healthcare resources and thus PPIs. Importantly, among five cohort studies, only two adjusted estimates for baseline differences and therefore potential confounders, including age and peptic ulcer history. Cardiovascular risk factors and socioeconomic status were not considered as potential confounders in any study.
Among the cohort studies, the pooled RR of the estimates from those with higher quality was considerably higher than that from those with lower quality (0.37 vs. 0.17). This implies that low-quality studies tended to publish results representing large beneficial effect of PPIs. In line with this, the funnel plot indicated an absence of small-size studies reporting small beneficial effect of PPIs. Despite the presence of heterogeneity and low number of studies, the direction of the effects in all studies and of the pooled effects in all subgroups is consistent (< 0.55), with all the upper bounds of the 95% CIs smaller than 1.0. These observations suggest a plausibly beneficial effect of co-therapy with PPIs on the risk of peptic ulcer in patients receiving long-term low-dose aspirin.
These findings on the impact of PPIs on peptic ulcer argue in favour of, but may be insufficient to justify the co-administration of PPIs. First, the clinical relevance of peptic ulcer prevention is unclear because peptic ulcer is a poor predictor of ulcer complications and can remain asymptomatic. In addition, only seven studies, of which five were observational, were included in the meta-analysis. In observational studies, it is difficult to fully prevent or account for confounding, and, unfortunately, most studies included in our analysis failed to appropriately adjust for confounders.
The current meta-analysis showed that concomitant use of PPIs in patients receiving low-dose aspirin corresponds to a 50% and 64% reduced risk of GI bleeding and UGIB respectively. Again, this finding should be interpreted with caution. There was methodological diversity in study design, and clinical heterogeneity in the treatments and baseline characteristics of the included patients. RCTs, case–control and cohort studies were all included in the meta-analysis. The inclusion criterion on the minimal exposure time of aspirin and PPIs differed between studies, ranging from 7 days to 6 months. The specific type of PPIs was not reported in more than half of the included studies, and differed between studies that report these data. Except for the study by Scheiman et al., which included only patients with a negative H. pylori test, none of the studies took the status of H. pylori infection into account. Helicobacter pylori infection may be a confounder, because it is associated with both the risk of UGIB and the use of PPIs in eradication therapy. Because the duration of PPI exposure was not reported in two observational studies, it is not clear whether PPI use in these studies was because of H. pylori eradication (short-term) or treatment of peptic ulcer disease (long-term). Among five-observational studies that reported on UGIB, only three studies adjusted the estimates for potential confounders, including age and peptic ulcer history. Importantly, no studies adjusted the estimates for other major potential confounders, such as H. pylori infection, CV risk factors and socioeconomic status.
As in the case of peptic ulcer, these findings on the impact of PPIs on the risk of GI events may not be sufficient to justify the use of PPIs in patients on long-term aspirin therapy. Only six studies reporting on UGIB, of which five were observational, were included in the meta-analysis. Although the findings overall indicate a beneficial effect of adding PPIs to aspirin on the risk of GI bleedings, the directions of the 95% CIs of the estimated RRs were not consistent across all studies. This observation, together with the evidence of publication bias, indicates that the magnitude of the beneficial effect of PPIs on GI bleeding remains uncertain.
Because only one study on the effect of PPIs on CV risk was found, we could not perform a meta-analysis of this effect. Hypothetically, PPIs might affect aspirin bioavailability by raising gastric pH and therefore attenuate the antiplatelet effect of aspirin. Only a few studies tested the latter hypothesis and showed contradictory results. The clinical significance of these hypotheses has recently also been seriously challenged. We found only one study that investigated this clinical significance. In this large retrospective cohort study, propensity scores were used to reduce bias in the estimated effect because of confounding factors. While residual confounding was considered unlikely, the authors acknowledged that it could not be excluded that the effect of PPIs might be confounded by a number of uncontrolled factors such as lipid levels, body mass index and smoking, on which there was no information. In addition, comorbidities that were not considered but might act as confounders in this study included severe reflux and Barrett's oesophagus; patients with these diseases are often prescribed PPIs and are at a high risk of CV death. Currently, the clinical impact of PPIs on the effect of aspirin on CV risk therefore remains uncertain.
Our study has some limitations. Because of the low number of the included studies, we pooled data from both RCTs and observational studies. Three case–control studies (reporting on UGIB) were not nested case–control studies and therefore we considered the odds ratio as RRs assuming that UGIB is a rare event. In four observational studies, neither relevant RRs nor relevant odds ratios were provided and we calculated the RRs using the reported number of events in the groups with and without a PPI therapy; in this way, it was not possible to adjust the RRs for differences in the patient baseline characteristics in these two groups.
In this study, we used the random-effects models to estimate the mean true effects in a hypothetical population of studies from which, the studies we found were assumed to be a random sample. Because the number of the included studies was relatively small, the estimates of the between-study variances (τ), and hence of the pooled effects and their confidence intervals, may not be precise when using the random-effects models. Additional meta-analyses using fixed-effects models showed that the pooled RRs obtained from the fixed-effects models (0.37 for peptic ulcer and 0.77 for GI bleeding) were slightly larger than those obtained from the random-effects models. The 95% CIs of the pooled RRs from the fixed-effects models [(0.30–0.45) for peptic ulcer and (0.68–0.89) for GI bleeding] were narrower, but both the lower and upper bounds were larger than those from the random-effects models. Thus, the random- and fixed-effects models produced similar results.
Despite the claim that inclusion of unpublished studies may help to avoid publication bias, we did not include unpublished studies because this inclusion can itself introduce bias for several reasons. First, the studies that can be retrieved may not be representative of all unpublished studies. Second, the quality of unpublished studies, given a lack of peer review, may be lower than that of published studies. Finally, data provided by the study investigators may not be complete and this may depend on whether or not the results are favourable. Besides the arguments above, there is also a practical reason to not include unpublished research. To ensure that the inclusion of unpublished studies does not cause bias, an exhaustive, manual search should be made to retrieve all relevant studies. In addition, a detailed and critical appraisal of the unpublished studies should be performed. There are numerous national and international registers and therefore a lot of time would be needed to search these sources, retrieve the documents and critically appraise the studies.
The current literature review showed that published studies on the impact of PPIs on the GI and CV outcomes in aspirin users were limited in both quantity and quality. Given the large number of patients concomitantly prescribed aspirin and PPIs, there is an urgent need for further studies on this topic. The clinical significance of peptic ulcer prevention in asymptomatic patients should be rigorously clarified. Also, more studies on the impact of PPIs on CV events and new studies on the effect of PPIs on aspirin adherence should be conducted. In addition, any possible long-term side effects of PPI co-therapy in patients receiving low-dose aspirin should be carefully reviewed and/or further investigated. A recent large study in 210 155 patients showed that the use of PPIs for two or more years was significantly associated with vitamin B12 deficiency. Long-term therapy with PPI might also be associated with increased risks of fracture, pneumonia, enteric infection and hypomagnesaemia, but more evidence is needed to confirm (or refute) these associations.
Since the risks of adverse GI events differ in various patient groups, studies should also focus on the benefit-risk ratios of co-administration of PPIs in particular patient groups with different risk factors. Given that long-term use of PPIs is costly, co-therapy with PPIs should be viewed also from a health economic perspective. For this, a pharmacoeconomic study is needed to provide comprehensive information on the clinical, quality of life and cost-consequences of co-prescribing PPIs for decision makers.
Discussion
Impact of PPIs on Peptic Ulcer
The current meta-analysis showed that concomitant PPI use in patients receiving low-dose aspirin confers a 73% reduced risk of peptic ulcer. However, the clinical significance of this effect is questionable. Peptic ulcer in all seven included studies were detected by endoscopy, but the associated symptoms were not reported in five of these studies. One study was conducted in asymptomatic patients and another reported that only 47% of patients had abdominal symptoms.
The pooled effect of PPIs on peptic ulcer should be interpreted with caution. The presence of significant heterogeneity indicates that this estimate may be biased. There was methodological diversity as we included both RCTs and cohort studies. Quality assessment of the studies revealed clinical heterogeneity in the treatments and baseline characteristics of the included patients. For example, the inclusion criterion on the minimal exposure time of the patients to aspirin and PPIs differed considerably between the observational studies, ranging from 7 days to 1 year for aspirin use and from 7 days to 3 months for PPI use. Types and doses of PPIs also differed between studies. One study selected only patients older than 65 years while the others included patients with different age ranges (min age, 18; max age, 92) and averages (67.4–79.7). The patient profiles of CV risk factors also varied across studies, partly because three studies did not exclude patients at very high CV risk while the others did. Age and the presence of CV risk factors may be confounders as they are likely to be associated with increased risks of GI events as well as with PPI use. Other potential confounders include a history of peptic ulcer disease and lower socioeconomic status; the latter is associated with Helicobacter pylori infection and the risk of peptic ulcer disease, and also with less access to healthcare resources and thus PPIs. Importantly, among five cohort studies, only two adjusted estimates for baseline differences and therefore potential confounders, including age and peptic ulcer history. Cardiovascular risk factors and socioeconomic status were not considered as potential confounders in any study.
Among the cohort studies, the pooled RR of the estimates from those with higher quality was considerably higher than that from those with lower quality (0.37 vs. 0.17). This implies that low-quality studies tended to publish results representing large beneficial effect of PPIs. In line with this, the funnel plot indicated an absence of small-size studies reporting small beneficial effect of PPIs. Despite the presence of heterogeneity and low number of studies, the direction of the effects in all studies and of the pooled effects in all subgroups is consistent (< 0.55), with all the upper bounds of the 95% CIs smaller than 1.0. These observations suggest a plausibly beneficial effect of co-therapy with PPIs on the risk of peptic ulcer in patients receiving long-term low-dose aspirin.
These findings on the impact of PPIs on peptic ulcer argue in favour of, but may be insufficient to justify the co-administration of PPIs. First, the clinical relevance of peptic ulcer prevention is unclear because peptic ulcer is a poor predictor of ulcer complications and can remain asymptomatic. In addition, only seven studies, of which five were observational, were included in the meta-analysis. In observational studies, it is difficult to fully prevent or account for confounding, and, unfortunately, most studies included in our analysis failed to appropriately adjust for confounders.
Impact of PPIs on Gastrointestinal Bleeding
The current meta-analysis showed that concomitant use of PPIs in patients receiving low-dose aspirin corresponds to a 50% and 64% reduced risk of GI bleeding and UGIB respectively. Again, this finding should be interpreted with caution. There was methodological diversity in study design, and clinical heterogeneity in the treatments and baseline characteristics of the included patients. RCTs, case–control and cohort studies were all included in the meta-analysis. The inclusion criterion on the minimal exposure time of aspirin and PPIs differed between studies, ranging from 7 days to 6 months. The specific type of PPIs was not reported in more than half of the included studies, and differed between studies that report these data. Except for the study by Scheiman et al., which included only patients with a negative H. pylori test, none of the studies took the status of H. pylori infection into account. Helicobacter pylori infection may be a confounder, because it is associated with both the risk of UGIB and the use of PPIs in eradication therapy. Because the duration of PPI exposure was not reported in two observational studies, it is not clear whether PPI use in these studies was because of H. pylori eradication (short-term) or treatment of peptic ulcer disease (long-term). Among five-observational studies that reported on UGIB, only three studies adjusted the estimates for potential confounders, including age and peptic ulcer history. Importantly, no studies adjusted the estimates for other major potential confounders, such as H. pylori infection, CV risk factors and socioeconomic status.
As in the case of peptic ulcer, these findings on the impact of PPIs on the risk of GI events may not be sufficient to justify the use of PPIs in patients on long-term aspirin therapy. Only six studies reporting on UGIB, of which five were observational, were included in the meta-analysis. Although the findings overall indicate a beneficial effect of adding PPIs to aspirin on the risk of GI bleedings, the directions of the 95% CIs of the estimated RRs were not consistent across all studies. This observation, together with the evidence of publication bias, indicates that the magnitude of the beneficial effect of PPIs on GI bleeding remains uncertain.
Impact of PPIs on Cardiovascular Events
Because only one study on the effect of PPIs on CV risk was found, we could not perform a meta-analysis of this effect. Hypothetically, PPIs might affect aspirin bioavailability by raising gastric pH and therefore attenuate the antiplatelet effect of aspirin. Only a few studies tested the latter hypothesis and showed contradictory results. The clinical significance of these hypotheses has recently also been seriously challenged. We found only one study that investigated this clinical significance. In this large retrospective cohort study, propensity scores were used to reduce bias in the estimated effect because of confounding factors. While residual confounding was considered unlikely, the authors acknowledged that it could not be excluded that the effect of PPIs might be confounded by a number of uncontrolled factors such as lipid levels, body mass index and smoking, on which there was no information. In addition, comorbidities that were not considered but might act as confounders in this study included severe reflux and Barrett's oesophagus; patients with these diseases are often prescribed PPIs and are at a high risk of CV death. Currently, the clinical impact of PPIs on the effect of aspirin on CV risk therefore remains uncertain.
Study Limitations and the Need for Further Research
Our study has some limitations. Because of the low number of the included studies, we pooled data from both RCTs and observational studies. Three case–control studies (reporting on UGIB) were not nested case–control studies and therefore we considered the odds ratio as RRs assuming that UGIB is a rare event. In four observational studies, neither relevant RRs nor relevant odds ratios were provided and we calculated the RRs using the reported number of events in the groups with and without a PPI therapy; in this way, it was not possible to adjust the RRs for differences in the patient baseline characteristics in these two groups.
In this study, we used the random-effects models to estimate the mean true effects in a hypothetical population of studies from which, the studies we found were assumed to be a random sample. Because the number of the included studies was relatively small, the estimates of the between-study variances (τ), and hence of the pooled effects and their confidence intervals, may not be precise when using the random-effects models. Additional meta-analyses using fixed-effects models showed that the pooled RRs obtained from the fixed-effects models (0.37 for peptic ulcer and 0.77 for GI bleeding) were slightly larger than those obtained from the random-effects models. The 95% CIs of the pooled RRs from the fixed-effects models [(0.30–0.45) for peptic ulcer and (0.68–0.89) for GI bleeding] were narrower, but both the lower and upper bounds were larger than those from the random-effects models. Thus, the random- and fixed-effects models produced similar results.
Despite the claim that inclusion of unpublished studies may help to avoid publication bias, we did not include unpublished studies because this inclusion can itself introduce bias for several reasons. First, the studies that can be retrieved may not be representative of all unpublished studies. Second, the quality of unpublished studies, given a lack of peer review, may be lower than that of published studies. Finally, data provided by the study investigators may not be complete and this may depend on whether or not the results are favourable. Besides the arguments above, there is also a practical reason to not include unpublished research. To ensure that the inclusion of unpublished studies does not cause bias, an exhaustive, manual search should be made to retrieve all relevant studies. In addition, a detailed and critical appraisal of the unpublished studies should be performed. There are numerous national and international registers and therefore a lot of time would be needed to search these sources, retrieve the documents and critically appraise the studies.
The current literature review showed that published studies on the impact of PPIs on the GI and CV outcomes in aspirin users were limited in both quantity and quality. Given the large number of patients concomitantly prescribed aspirin and PPIs, there is an urgent need for further studies on this topic. The clinical significance of peptic ulcer prevention in asymptomatic patients should be rigorously clarified. Also, more studies on the impact of PPIs on CV events and new studies on the effect of PPIs on aspirin adherence should be conducted. In addition, any possible long-term side effects of PPI co-therapy in patients receiving low-dose aspirin should be carefully reviewed and/or further investigated. A recent large study in 210 155 patients showed that the use of PPIs for two or more years was significantly associated with vitamin B12 deficiency. Long-term therapy with PPI might also be associated with increased risks of fracture, pneumonia, enteric infection and hypomagnesaemia, but more evidence is needed to confirm (or refute) these associations.
Since the risks of adverse GI events differ in various patient groups, studies should also focus on the benefit-risk ratios of co-administration of PPIs in particular patient groups with different risk factors. Given that long-term use of PPIs is costly, co-therapy with PPIs should be viewed also from a health economic perspective. For this, a pharmacoeconomic study is needed to provide comprehensive information on the clinical, quality of life and cost-consequences of co-prescribing PPIs for decision makers.