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


An umbrella review was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) checklist.[17] A protocol was registered with PROSPERO (CRD42019148366).

Search Strategy

The search strategy (see Supplementary Material) was developed by a medical information specialist in consultation with the review team, and peer reviewed prior to execution using the PRESS Checklist.[18] Using the OVID platform, we searched MEDLINE and Embase on 10 July 2019. Strategies utilised controlled vocabulary (eg, 'Proton Pump Inhibitors') and free text terms (eg, 'Nexium', 'Prevacid'). Amended versions of search filters developed by CADTH[19] were used, with bespoke free text and controlled vocabulary terms added for known or suspected adverse effects of PPIs. Searches were limited by date from 2009 and no language limits were applied.

Eligibility Criteria

Given the volume of systematic reviews retrieved, articles published before 2015 were excluded. Only published systematic reviews with meta-analyses, which clearly identified PPIs as the intervention of interest, as compared to a non-PPI group, and which investigated associations of any adverse events were included. With respect to population treated, studies assessing broad populations at risk for gastrointestinal disorders (ie, dyspepsia and GORD), were considered eligible.

Exclusion criteria included: paediatric populations; studies focused on specific non-reflux disease areas or specific patient groups (eg, intensive care patients, stroke patients); combination of PPIs with other therapies (eg, clopidogrel); comparison of PPIs to a defined drug or drug class; investigation of outcomes not identified as adverse, and qualitative systematic reviews. For completeness, any effect estimates (ie, risk ratio [RR], odds ratio [OR], hazard ratio [HR]) were accepted. Effect estimates were not directly compared; rather, only their significance and direction were considered.

Study Selection and Data Extraction

Two authors (SC and ES) independently screened all records using DistillerSR (Evidence Partners Inc), and resolved conflicts by discussion. Relevant data were extracted from included articles by one reviewer (SC) and peer reviewed by a second reviewer (ES). Extracted details included study characteristics (eg, search details, study eligibility criteria), meta-analysis results, study quality assessments and study limitations. A standardised electronic form (Excel; Microsoft) was used to ensure consistency.

Where adverse outcomes were investigated in only one meta-analysis, these studies were automatically selected for presentation of results and discussion.[20–29] Where multiple meta-analyses investigated the same outcome, the most comprehensive one was selected for discussion to avoid double-counting results, enhance rigour of the assessment and simplify the volume of results for interpretation. This practice has been used in other published umbrella reviews when overlapping meta-analyses exist.[18,19] Key determinants for selection included: number of patients, number of included studies, methodological quality based on AMSTAR 2 assessment (ie, ratings of critically low were not eligible unless no report of higher quality was retrieved) and publication date. In all cases, rationale for selection was recorded. The magnitude and significance of associations across meta-analyses were qualitatively assessed and compared in a structured manner through use of disease categorical groupings, plots of sample size relative to effect size and recording of statistical significance.

For all outcomes, if selected meta-analyses reported differential findings in statistical significance compared with non-selected meta-analyses, discrepancies were noted. Identified adverse outcomes were grouped based on organ system or disease area. Where identified, subgroup analyses of PPI duration of use were also discussed (regardless of duration cut-offs selected by authors). As with broader adverse outcome associations, subgroup analyses from publications considered the most comprehensive were selected for discussion.

Study Quality Assessment

Methodological quality of included systematic reviews was assessed using AMSTAR 2 (Assessing the Methodological Quality of Systematic Reviews).[30] The AMSTAR 2 assessment contains 16 items and generates an overall quality rating based on weaknesses in critical domains. Assessments were peer reviewed among the authors. Additionally, author-reported quality assessments were compiled. Where available, Newcastle-Ottawa Scale (NOS) scores assessing quality of supporting observational studies within the systematic reviews were extracted and averaged. Modified versions of the scale were also considered; however, total scores were converted to a 9-point scale. Finally, GRADE assessments were conducted for all included systematic reviews with meta-analyses, to assess the certainty of evidence.[31] If possible, author-reported GRADE assessments were used.