Umbrella Review of 42 Systematic Reviews With Meta-analyses

The Safety of Proton Pump Inhibitors

Elizabeth M. Salvo; Nicole C. Ferko; Sarah B. Cash; Ailish Gonzalez; Peter J. Kahrilas


Aliment Pharmacol Ther. 2021;54(2):129-143. 

In This Article


Forty-two systematic reviews with meta-analyses were identified which fulfilled the eligibility criteria. Collectively, these studies investigated a range of adverse events, including but not limited to fracture, kidney disease, Clostridium difficile infection, pneumonia, cancer and cardiovascular events. The study flow is depicted according to PRISMA guidelines in Figure 1.[17] The included reviews varied with respect to their search strategies (eg, databases searched), eligibility criteria for study inclusion (eg, study designs were restricted to randomised controlled trials in some cases) and reporting style (ie, types of effect estimates reported; Table S2). The majority of reviews included observational studies, with few including randomised data (Table 1).

Figure 1.

PRISMA diagram for selection of systematic reviews with meta-analyses

The included systematic reviews with meta-analyses were divided into categories based on organ system or disease area investigated, with some studies assessing more than one outcome type (Figure 2). These outcomes included (n = number of meta-analyses): bone-related events (n = 9),[20,25,32–38] kidney-related events (n = 8),[25,39–45] infections (n = 11),[21,22,25,46–53] cardiovascular diseases and related events (n = 4),[23,24,54,55] gastrointestinal-related diseases (n = 7),[26–28,56–59] cancer outcomes (n = 3),[25,57,60] and neurological outcomes (n = 4).[29,61–63] Sample sizes varied widely across the included meta-analyses, from 475 patients in the study of a colitis outcome by Tong et al[26] to 7.6 million patients in the study of pneumonia by Wang et al.[53] Most meta-analyses assessed over 100 000 patients and several included at least 1 million patients. Table 1 summarises details of the included studies.

Figure 2.

Systematic reviews with meta-analyses included in the umbrella review which investigated the association of adverse events with PPI use. Several publications reported meta-analyses for more than one adverse outcome

Bone-related Outcomes

Nine systematic reviews focused on bone-related outcomes, most of which investigated associations of PPI use with fractures (hip fracture,[25,32–38] spine fracture[32,33,35,36,38] and any-site fracture[32,33,35,36,38]). The most comprehensive analyses of hip and any-site fractures included 1.5 million and 2.1 million patients respectively, and significant associations were determined (Figure 3A).[35,37] The most comprehensive analysis of spine fracture included far fewer patients but still reported a significant association with PPI use.[35] The same authors determined a significant association for osteoporosis. Wrist fracture was also investigated in one study; however, the association was not significant.[32] Finally, Chappuis et al determined a significant association for dental implant failure.[20] All of the selected meta-analyses of bone outcomes, with the exception of dental implant failure, were informed by observational data. Nassar et al, which was not selected as the most comprehensive meta-analysis, reported no association of PPI use with spine fracture.[36]

Figure 3.

Summary of evidence for the association of PPI use with adverse outcomes in systematic reviews with meta-analyses categorised as the most comprehensive for each outcome. Bar charts are presented based on outcome type: (A) bone-related outcomes, (B) kidney-related outcomes, (C) infection outcomes, (D) cardiovascular and mortality outcomes, (E) cancer, (F) gastrointestinal outcomes, (G) neurological outcomes. Data are based on observational studies, unless otherwise noted. *Associations found to be statistically significant for PPI use and adverse outcome (P < 0.05)

Kidney Outcomes

Adverse kidney outcomes were investigated in eight meta-analyses, including chronic kidney disease (CKD),[25,39–44] ESRD,[39–44] AKI,[40,41,44,45] and acute interstitial nephritis (AIN). The most comprehensive meta-analyses determined a significant association with PPI use for each of these outcomes (Figure 3B).[39,44] Notably, Wu et al included almost 2.4 million patients in their analysis of AKI.[44] All of the selected meta-analyses of adverse kidney outcomes were informed by observational data. Meta-analyses not selected as the most comprehensive concluded similar findings in all cases, indicating broad alignment across multiple investigations.

Infection Outcomes

Eleven studies investigated infection outcomes.[21,22,25,46–53] Associations of C. difficile[21,46–51] and pneumonia[25,52,53] infections with PPI use were frequently investigated in the included meta-analyses, and a single meta-analysis investigated associations with various enteric infections.[22]

In the meta-analyses selected as most comprehensive, effect estimates were statistically significant in the cases of C. difficile infection (both incident[51] and recurrent infections[21]), pneumonia,[52] as well as enteric infections (Figure 3C).[22] Hafiz et al reported a statistically significant OR for the association between enteric infections and PPI use that exceeded 4.0.[22] Importantly, more than 50 studies were included in each of the selected pneumonia[52] and C. difficile[50] analyses.[51,53] Furthermore, the selected meta-analysis for the pneumonia outcome included a total of 7.6 million patients and was informed by a combination of RCT and observational evidence (10 RCTs and 48 observational studies).[53]

Although the Trifan et al study (selected as most comprehensive for C. difficile infection) pooled all acquisition modes,[51] another meta-analysis investigated hospital-acquired and community-acquired C. difficile separately,[47] and similarly there were meta-analyses which considered only hospital-acquired[46,50] or only community-acquired infections.[25,49] For the pneumonia outcome, the Wang et al study (selected as most comprehensive) also pooled hospital and community acquisitions,[53] while other identified meta-analyses considered only community acquisition.[25,52] Regardless of acquisition mode, a significant association with PPI use was reported for all identified infection outcomes, across most of the included meta-analyses. Exceptions include a meta-analysis conducted by Zacharioudakis et al investigating hospital-acquired C. difficile, and a meta-analysis conducted by Cao et al investigating community-acquired C. difficile, neither of which reported significance.

Cardiovascular Events and Mortality

Four meta-analyses investigated cardiovascular outcomes, with mixed findings.[23,24,54,55] The most comprehensive of these was supported entirely by RCT evidence (7540 patients) in the GORD indication, and reported a significant association for a range of cardiovascular events with PPI use (Figure 3D).[55] Alternatively, Farhat et al conducted more specific meta-analyses for stroke and myocardial infarction events, neither of which were significant.[23] Finally, Shiraev et al investigated all-cause mortality, and concluded that a significant association exists with PPI use, using observational evidence.[24]


A significant association of PPI use with gastric cancer was reported by Islam et al in a meta-analysis including roughly 18 000 patients.[25] This conclusion is aligned with two other meta-analyses.[57,60] Conversely, the presence of an association between PPI use and colorectal/pancreatic cancers was also investigated by Islam et al; however, these findings were not statistically significant (Figure 3E).[25] All included meta-analyses of cancer outcomes were informed by observational studies.

Gastrointestinal Outcomes

The association of PPI use with several non-infectious gastrointestinal conditions has also been investigated in the literature and seven meta-analyses were identified which assessed multiple unique outcomes. Single meta-analyses investigated associations with gastric mucosal atrophy,[28] colitis[26] and small intestinal bacterial overgrowth,[27] and two meta-analyses each were identified investigating associations with hypomagnesaemia[58,59] and fundic gland polyps.[56,57] A significant association with PPI use was concluded for all of these outcomes except mucosal atrophy (Figure 3F), and similar meta-analyses for hypomagnesaemia and fundic gland polyps outcomes were aligned. All identified meta-analyses investigating these outcomes were informed by observational studies, with exception of the recent meta-analysis by Martin et al which included a single RCT in addition to 19 observational studies.[56]

Neurological Outcomes

Three included studies investigated associations between dementia and/or Alzheimer's disease and PPI use, all of which were not statistically significant (Figure 3G).[61–63] Alternatively, one meta-analysis did identify a significant association of PPI use with fall risk.[29] These meta-analyses were supported solely by observational evidence; however, those selected as the most comprehensive included >360 000 patients.

Subgroup Analyses: Duration of PPI use

Several meta-analyses reported subgroup analyses investigating duration of PPI use (Figure 4). Non-statistically significant increases in effect estimates with longer durations of use were reported in several of these analyses (hip fracture,[37] osteoporosis,[33] chronic kidney disease,[39] pneumonia,[53] gastric cancer,[57] cardiovascular events[55]) and there were two cases where these increases were statistically significant: end-stage renal disease[42] and fundic gland polyps.[57] In the case of end-stage renal disease, the size of the association increased significantly between durations of 1–3 months and 6–12 months, and again for durations of 12–24 months. In the case of fundic gland polyps, the size of the association increased significantly when duration of use exceeded 12 months.[57]

Figure 4.

Subgroup analyses of PPI duration of use. The majority of subgroup analysis identified reported a risk ratio. Those which reported odds ratio (†) or effect estimate (††) are indicated. *Associations are found to be statistically significant for PPI use and adverse outcome (P < 0.05)

Study Quality and Certainty of Evidence

AMSTAR 2 ratings for the included systematic reviews with meta-analyses were mixed. Sixteen of 42 meta-analyses were assigned a critically low score, whereas 11 were assigned a moderate score (Table 1). Full assessment details for all included studies are available in Table S4.

Additionally, use of the NOS scale to ascertain quality of primary studies was reported by authors for 32 of the 42 included systematic reviews (Table S3). Of these, six studies reported use of a modified NOS scale (two of which did not report numeric data for compilation,[25,52] and one of which was not considered given uncertainty in the version used[43]). Taken together, mean NOS scores were calculated for 29 of the 42 included systematic reviews (69%) (Figure 5). Therefore, across observational studies included within the meta-analyses, mean NOS scores were greater than 7 of 9 points (ie, a good/high quality score) for 20 of the 29 (69%) meta-analyses considered, and varied from 5.64 to 8.83 out of 9 points.

Figure 5.

Comparison of mean Newcastle-Ottawa Scale scores across meta-analyses selected as most comprehensive. *Sun et al (2017) only included RCTs in their meta-analysis, and as such use the NOS instrument was not applicable. A modified version of the NOS scale was used, but scoring data were unavailable. In some instances, modified Newcastle-Ottawa Scale scores were converted to scores out of 9 points (see Supplementary Material)

Individual systematic reviews with meta-analyses were usually considered to have very low certainty of evidence in the GRADE assessment, due to the incorporation of predominantly observational evidence. High levels of heterogeneity or a lack of investigation of the impact of low study quality on results were often seen as rationale for downgrading certainty levels. Studies were very rarely downgraded based on indirectness or potential for publication bias.