Evidence based on millions of patients with acid-related disorders, mainly derived from real-world observational data, supports the association of PPI use with a range of adverse outcomes. Specifically, our umbrella review demonstrated that PPIs were associated with increased adverse outcomes related to the bones,[20,25,32–38] kidneys,[25,39–45] gastrointestinal tract conditions,[26–28,56–59] infections,[21,22,25,46–53] and falls compared to no PPI use. Investigations of the association between cardiovascular events and PPI use were mixed; however, a significant association was found for all-cause mortality. Across meta-analyses selected as being most comprehensive, there was a greater than twofold increased likelihood of the following outcomes with PPI use: dental implant failure, enteric infections, recurrent C. difficile, colitis, gastric cancer, colitis and fundic gland polyps. Furthermore, there were two outcomes for which a duration of use relationship was identified with PPIs: end-stage renal disease and fundic gland polyps. However, it must be noted that these findings were mostly supported by meta-analyses of observational studies and therefore only association, rather than causation, can be established. Furthermore, despite the statistical significance of several identified associations, the effect estimates were generally small and further study is required to determine whether or not weak associations will persist with increased evidence. Eight of the 42 studies provided adequate information to calculate the number needed to harm for one or more adverse outcomes, which ranged from 7 to 240 patients across several distinct outcomes. This finding suggests that most identified adverse outcomes were generally rare.
Our umbrella review also revealed that there were no statistically significant associations observed between PPI use and non-gastric cancers, including pancreatic cancer and colorectal cancer. Furthermore, although our analysis revealed an association between PPI use and fall risk, there were no significant associations demonstrated between PPI use and dementia,[61–63] or Alzheimer's disease.[61,62] Most of these absences of association were supported by meta-analyses that included more than 100 000 patients.
Although evidence indicates that PPIs are effective therapies for acid-related diseases (predominantly GORD), 25%-70% of PPI prescriptions are considered inappropriate.[6,7] Inappropriate prescriptions may refer to PPI use for patients without formal diagnoses as well as continued prescription of PPIs to patients following discharge from hospital settings. This chronic, widespread PPI use increases the risk of adverse events, even rare adverse events.[6,7] To reduce unnecessary risk, guidelines have been developed for cautiously deprescribing PPIs.[6,7,64] Regardless of the estimated size of the risk or the likelihood of causation, risk will always dominate a risk-benefit assessment in the absence of benefit. However, it is also important that clinicians critically evaluate decisions to restrict PPI prescriptions—PPIs are the most effective options for certain patients, especially those with severe acid-related disease. We hope that this paper brings focus to this issue, as PPIs are being widely prescribed in primary care and the indications for prescription are inconsistent.
Possible Mechanisms for Adverse Events Associated With PPI use
The mechanism of action of PPIs is important when considering potential PPI-related adverse events. PPIs selectively bind and inactivate hydrogen/potassium ATPase, resulting in profound inhibition of hydrochloric acid release from parietal cells and a sustained increase in gastric pH. This increased gastric pH is known to impair vitamin B12 absorption. Prolonged B12 deficiency may result in ataxia, which may increase risk of falls. It has also been suggested that calcium absorption may be impaired linking PPI use to osteoporosis and fractures.[4,20,37,66] Magnesium transport is also impacted in altered pH conditions[58,67–69] and hypomagnesaemia is thought to induce kidney-related adverse outcomes due to endothelial cell dysfunction, inflammation, reduced glomerular filtration and oxidative stress.[41,44] PPI metabolite deposition within the renal tubule-interstitium may induce acute interstitial nephritis and acute kidney injury.[39,44,45] More simply, increased rates of various infections have been associated with PPI use, as reduced gastric acidity creates an opportunity for ingested microorganisms to remain virulent (C. difficile, gastrointestinal pathogens, bacteria potentially causing pneumonia).[22,47,52] Cardiovascular homoeostasis may be affected by PPI-induced changes in vasodilation and platelet aggregation, in addition to hypomagnesaemia. Finally, the association of PPI use with gastric cancer is proposed to result both from accelerating the progression of H. pylori induced atrophic gastritis and by increased gastrin levels promoting enterochromaffin-like cell hyperplasia. However, apart from the increased occurrence of enteric infections, these proposed mechanisms for PPI-related adverse events are not supported with clinical evidence leaving the issue of causation unresolved and an appropriate focus of future research.
Other Investigations of Adverse Events Associated With PPI use
A previous umbrella review of the association of PPI use with adverse outcomes was published by Abramowitz et al. However, the eligibility criteria of our review were broader than in that review (limited to the GORD indication) and more outcomes were captured herein. Their study reported associations between PPI use and fracture, pneumonia and enteric infections.
Outside of umbrella reviews and meta-analyses, a recent large US-based longitudinal observational study examined mortality in a database of more than 6 000 000 US veterans over 5.7 years. The cohort study found that PPI use was associated with a small, statistically significant excess of cause-specific mortality including death due to cardiovascular disease, chronic kidney disease and upper gastrointestinal cancer. There were 45.2 thousand excess deaths among PPI users and the risk increased with more prolonged use. Strengths of this observational study included its sample size, prospective design and the use of propensity-matched scoring to control for confounders. Also noteworthy, a very recent US population-based survey indicated a dose-dependent association between PPI use and risk of COVID-19 virus positivity. Given the recent publication dates, these studies were not captured in the meta-analyses included within our review.
Most notably, a recently published trial by Moayyedi et al contradicted the findings of this umbrella review. In that placebo-controlled randomised trial of atherosclerotic vascular disease treatments, 17 598 patients were further randomised 1:1 to either PPI (pantoprazole) or placebo and the occurrence of a priori specified adverse events (pneumonia, C. difficile infection, enteric infections, fractures, gastric atrophy, chronic kidney disease, diabetes, chronic obstructive lung disease, dementia, cardiovascular disease, cancer, hospitalisations and all-cause mortality) was tracked for an average of 3 years. PPI use was not associated with any of these adverse events with the exception of enteric infections. Owing to both the recent publication date and the focused patient population (cardiovascular disease), that study was not captured by any meta-analyses identified herein. The prospective Canadian study CaMos is also relevant to this discussion wherein bone mineral density (the proposed mechanism through which fracture risk could be increased) was monitored in patients receiving PPIs. That study found no association between PPI use and bone demineralisation over 10 years, emphasising the importance of investigating putative mechanisms of adverse events.
Strengths and Limitations
The comprehensive nature of this review is unique given the breadth of adverse outcomes that were collectively assessed and summarised. The rationale for selection of the most comprehensive meta-analysis for each adverse event ensured that the discussion was up-to-date and not redundant (in most cases, it also resulted in a larger number of patients informing the discussion). Additionally, both the methodological quality of included meta-analyses as well as the quality of primary studies included therein were considered by conducting AMSTAR 2 assessments and by compiling available NOS results of primary studies respectively. This approach was undertaken because of the preponderance of observational evidence and the high frequency of NOS reporting. The broad nature of the eligibility criteria applied in this review also ensured that the investigation was comprehensive. Although some studies were excluded due to a focus on specific populations, the use of broad criteria should ensure that these conclusions are generally applicable across a range of patients using PPIs. Finally, the search strategy used was optimised based on recent guidance; adverse event filters validated by CADTH were used in conjunction with several keywords to ensure that the maximum number of adverse outcome reports was likely to be identified.[19,75]
There were also limitations to this review. Most importantly, the large amount of observational evidence increases the risk of unrecognised bias. Although these primary studies may be considered moderate to high quality based on author-reported NOS evaluations, these assessments are within the constraints of maximum quality achievable for an observational study. Inherently, the GRADE approach automatically considers observational evidence to have low certainty (as compared with randomised trials). This is because of unavoidable risks of bias (eg, lack of randomisation, lack of blinding, unknown or inaccurately accounted for confounding variables). Several of the included meta-analysis estimates had high heterogeneity based on author-reported I2 values, which further impacted GRADE assessments of certainty (Table S1). Furthermore, despite its widespread use, the NOS tool has been critiqued for low inter-rater reliability, partly due to lack of guidance. The ROBINS-I is another tool used to assess the risk of bias in observational studies, scoring them for heterogeneity, which is fundamentally different from the NOS tool. However, ROBINS-I assessments were not reported in any of the studies included herein. It should be noted that not all observational data are fatally flawed, and study quality assessment of observational data is still essential. Future observational research in this area should emphasise rigorous methods, such as propensity score matching, to help mitigate bias. Finally, the AMSTAR 2 assessments also identified areas for improvement in this dataset. Particularly, the studies assigned 'critically low' scores did not adequately assess risk of bias or account for potential bias in their interpretations.
A widely touted benefit of observational studies is the potential for including very large sample sizes, powering them to detect rare events.[13–15] Even so, they only measure association, indicating that a risk factor and an adverse event occur together; this does not establish that the risk factor causes the adverse event. Alternatively, associations can result from chance, bias and/or confounding variables. This is particularly relevant when there is no proven mechanistic hypothesis, as is the case with several of the adverse events considered here: ice cream consumption is associated with warm weather, as is gun violence. Consequently, ice cream consumption is associated with gun violence. However, restricting ice cream consumption is unlikely to impact on gun violence.
Finally, it is possible that additional eligible studies of higher quality may have been identified by broadening the search date restriction or by searching additional databases. However, the restriction of study publication date was meant to ensure the most current evidence was incorporated with minimal duplication of study inclusion.
Aliment Pharmacol Ther. 2021;54(2):129-143. © 2021 Blackwell Publishing