Meta-analysis: proton pump inhibitor use and the risk of community-acquired pneumonia

Authors


Dr J. Johnstone, Department of Medicine, McMaster University, 3200 MDCL, 1200 Main Street W, Hamilton, ON, Canada.
E-mail: johnsj48@mcmaster.ca

Abstract

Aliment Pharmacol Ther31, 1165–1177

Summary

Background  Observational studies examining the association between proton pump inhibitor (PPI) use and risk of community-acquired pneumonia are conflicting.

Aim  To assess systematically the association between risk of community-acquired pneumonia and PPI use in adults.

Methods  We searched MEDLINE, EMBASE and CINAHL databases between 1988 and January 2010. Two reviewers independently selected studies based on eligibility criteria and extracted data. Included studies evaluated adults (≥18 years) who took PPIs as an out-patient. The primary outcome was community-acquired pneumonia. Only observational studies with a comparison arm were included.

Results  Over 2600 citations were reviewed. Six studies were included. All were nested case-control studies. Meta-analysis found an increased risk of community-acquired pneumonia associated with PPI use [OR 1.36 (95% CI 1.12–1.65)]; significant heterogeneity remained (I2 92%, P < 0.001). In exploratory subgroup analysis, short duration of use was associated with an increased odds of community-acquired pneumonia [OR 1.92 (95% CI 1.40–2.63), I2 75%, P = 0.003], whereas chronic use was not [OR 1.11 (95% CI 0.90–1.38), I2 91%, P < 0.001], a significant interaction (P < 0.005).

Conclusions  Heterogeneity precluded interpretation of the summary statistic. Exploratory analysis revealed that duration of PPI use may impact the risk of community-acquired pneumonia, a finding that should be explored in future studies.

Introduction

Proton pump inhibitors were introduced into clinical practice in 1988 and have since become the mainstay of therapy for many acid-related gastrointestinal disorders including peptic ulcer disease,1, 2 gastro-oesophageal reflux disease (GERD)3–5 and non-ulcer dyspepsia.6, 7 Proton pump inhibitors are among the top five most prescribed drugs in North America,8 which is, in part, due to their safety profile.9 They are generally well tolerated and serious adverse events are rare; however, there have been an increasing number of reports linking the chronic use of proton pump inhibitors to significant adverse effects including drug-drug interactions,10–12Clostridium difficile associated diarrhoea,13 hip fractures14 and community-acquired pneumonia.15–21 Results of observational studies describing the association between proton pump inhibitors and risk of community-acquired pneumonia have been conflicting.15–21

A recent systematic review and meta-analysis of randomized controlled trials of proton pump inhibitors examined the association between their use and respiratory tract infections.22 In the seven trials that fulfilled inclusion criteria, the reported prevalence of respiratory infections was low (4% overall; odds ratio 1.42, 95% CI 0.86–2.35) and the incidence of pneumonia was not specified. A separate review attempted to estimate the association of esomeprazole and community-acquired pneumonia by pooling data from an adverse events database of 31 clinical trials.23 In total, there were 15/6534 (0.22%) cases of pneumonia in the esomeprazole group compared to 7/3358 (0.21%) in the placebo arm [RR 0.94 (99% CI 0.29–3.07)]; however, the low event rate precluded definitive conclusions.

At present, there are insufficient clinical trial data to determine whether an association between proton pump inhibitors and community-acquired pneumonia exists. Observational studies, which can evaluate rare outcomes, harms and outcomes not previously known to be important, are well-suited to address this clinical question.24 As such, we systematically evaluated observational studies to determine the risk of community-acquired pneumonia in adults associated with out-patient proton pump inhibitor use when compared with non-users of proton pump inhibitors.

Methods

All observational studies evaluating the association between out-patient use of proton pump inhibitor therapy and the first episode of community-acquired pneumonia following cohort enrolment were eligible for inclusion (Table 1). Observational studies with a comparator arm were included, as were all doses, frequency and duration of proton pump inhibitor therapy. The comparator arm included participants either not on gastric acid suppression therapy or on non-proton pump inhibitor acid suppression therapy.

Table 1.   Study inclusion and exclusion criteria
  1. CAP, community-acquired pneumonia; HIV, human immunodeficiency virus.

  2. * Nosocomial pneumonia includes Ventilation-associated pneumonia (VAP), pneumonia that arises after 48–72 h of mechanical ventilation; Hospital-associated pneumonia (HAP), pneumonia that occurs 48 h or more after admission to hospital; Health care-associated pneumonia (HCAP), pneumonia in any patient who was hospitalized in an acute care hospital for two or more days within 90 days of the pneumonia, resides in a nursing home or long-term care facility or spent any time receiving care within a hospital within 30 days of the pneumonia.27

Inclusion criteria
PopulationAdults (≥18 years)
InterventionOut-patient exposure to a proton pump inhibitor including:
• Ompeprazole
• Lansoprazole
• Esomeprazole
• Pantoprazole
• Rabeprazole
OutcomeFirst episode of community-acquired pneumonia following cohort enrolment
MethodsObservational studies with a control arm including:
• Cohort studies
• Case-control studies
• Cross sectional studies
Exclusion criteria
PopulationPaediatric population (<18 years)
Individuals at increased risk of aspiration pneumonitis including:
• Perioperative patients
• Patients with gastric feeding tubes
• Nursing home residents
Hospitalized adults
Immunocompromised patients including:
• Patients who have had an organ or stem cell transplant
• Patients with cancer receiving active therapy
• Patients with HIV and CD4 counts <200/μL
InterventionIn-hospital use of proton pump inhibitors including:
• In-hospital therapy for acute gastrointestinal bleed
• In-hospital stress ulcer prophylaxis
• Perioperative aspiration pneumonitis prophylaxis
Administration of proton pump inhibitors as part of therapy for H. pylori
ComparisonAlternative proton pump inhibitor or dose of proton pump inhibitor
OutcomeNon-CAP outcomes including:
• Aspiration pneumonitis
• Aspiration pneumonia
• Nosocomial pneumonia*
MethodsRandomized controlled trials
Case reports and case series
Reports not involving primary data including:
• Narrative review
• Systematic review
• Meta-analysis
• News items
• Consensus statements
• Guidelines
• Opinion pieces/editorials
Companion articles of original studies

We excluded children (<18 years) as there are few indications for proton pump inhibitor use in children and accordingly, few out-patient prescriptions.25 We also excluded in-patient proton pump inhibitor therapy use, as hospitalization predisposes individuals to hospital acquired pneumonia, a different class of pneumonia.26, 27 Cohorts comprised entirely of immunocompromised patients (i.e. >20 mg prednisone/day for >21 days, other immunosuppressive agents, cancer with recent chemotherapy, or HIV with CD4 count <200/μL) were excluded, as this group is at risk of infection with respiratory pathogens beyond the typical spectrum of community-acquired pneumonia organisms.28 Studies evaluating use of proton pump inhibitor therapy for Helicobacter pylori were excluded as this treatment also includes the use of antibiotics with activity against respiratory pathogens.29 Finally, patients at increased risk of aspiration pneumonitis (i.e. perioperative patients, patients with gastric feeding tubes and nursing home residents) were excluded in an attempt to minimize misclassification bias.

We performed an electronic search in MEDLINE (1950–8 January 2010), EMBASE (1980–week 53, 2009) and CINAHL (1982–8 January 2010) using the following terms: anti-ulcer agent, proton pump inhibitor, PPI, omeprazole, lansoprazole, esomeprazole, pantoprazole, rabeprazole, antacid, pneumonia, community-acquired pneumonia and CAP (see supplementary data Appendix S1 for our full search strategy). We limited our search to studies after 1988 as proton pump inhibitors were first introduced that year. To identify relevant citations further, we searched article reference lists and used the PubMed ‘related articles’ feature. Finally, we contacted a content expert to identify any study missed by the electronic searches.

Independently and in duplicate (JJ and KN), all titles and abstracts were screened for eligibility based on the above eligibility criteria. All potentially relevant articles were reviewed independently in full by both authors for eligibility. Agreement was measured using a weighted Kappa score.30 Disagreement among reviewers was resolved by consensus. In situations where the reviewers were unsure of study eligibility, the corresponding authors of the studies were contacted for missing information. Once additional information was obtained from the authors, the two reviewers re-assessed eligibility. Data collection for all information on the study population, intervention, outcome and methods was also performed independently and in duplicate to minimize abstraction errors.31

The primary outcome of interest was risk of community-acquired pneumonia. We used the published adjusted odds ratio and corresponding 95% confidence intervals from each included study to determine the association between proton pump inhibitor use and the outcome of interest. Pooling of the log transformed odds ratios using the inverse variance method was performed in Review Manager 5.0 (Cochrane Collaboration, Copenhagen, Denmark) to determine summary measures of risk using a random effects model.32 We justified pooling all types of proton pump inhibitors as they have similar efficacy and are generally used interchangeably;33, 34 however, only results of similar study designs were pooled because of the potential varying risk of bias inherent to differing study designs.35

We evaluated heterogeneity using the χ2 and I2 statistic.24 We defined significant heterogeneity as a χ2 of <0.10 or an I2 statistic of >50%. Although there were many potential reasons for heterogeneity (Table 2), we considered three a priori hypotheses to explain potential variability between studies: (i) severity of pneumonia (i.e. determined by need for admission to hospital vs. no admission to hospital; (ii) duration of proton pump inhibitor therapy [i.e. recently prescribed (<30 days) vs. chronic user (≥30 days)]; and (iii) dosage of proton pump inhibitor therapy [i.e. low dose (<1 defined daily dose) vs. high dose (>1 defined daily dose)]. We also performed tests of interaction for these subgroups.36

Table 2.   Potential sources of heterogeneity
Population
 Age
 Gender
 Co-morbidity
Intervention
 Dose of proton pump inhibitor
 Duration of use of proton pump inhibitor
 Type of proton pump inhibitor
 Reason for proton pump inhibitor use
Comparison
 No anti-ulcer agent
 Alternative anti-ulcer agent
Outcomes
 Severity of community-acquired pneumonia
Methodology
 Method of obtaining exposure information
 Method of obtaining outcome information
 Risk of bias
  Cohort vs. case control vs. cross-sectional
  Adequacy of adjustment of confounding variables
  Attempts to exclude aspiration pneumonitis

Two a priori sensitivity analyses were planned for the primary outcome: (i) we excluded studies that defined pneumonia outcomes using only database coding, given the potential misclassification bias attributed to the non-specific symptoms and signs of community-acquired pneumonia;26, 37 and (ii) we excluded studies that did not adequately adjust for potential confounding variables known to increase the risk of community-acquired pneumonia including age, gender, smoking status, COPD, history of stroke and functional status.38, 39 GERD was also included as a potential confounder, as it is a known cause of aspiration pneumonitis,40 which can be misclassified as community-acquired pneumonia.

Reporting bias was to be assessed by Funnel plots, if more than 10 studies were included in the analysis of the primary outcome.24

Results

The electronic search yielded 2616 citations (see supplementary data). The content expert identified one additional published citation,20 not yet indexed in any of the electronic databases. Manual review of the reference lists did not yield any additional citations. After 140 duplicates were excluded, 2477 titles and abstracts were reviewed. As outlined in Figure 1, 2383 citations did not meet the eligibility criteria; therefore, full text review was performed on 94 manuscripts. Of these, 82 were excluded leaving 12 potential articles for inclusion.15–21, 41–45 Six of these 12 articles required further information before definitive conclusions could be made.21, 41–45 Once additional information was obtained, all six studies were excluded for the following reasons: three studies provided only adjusted odds ratios for ‘gastric acid suppression therapy’ and risk of community-acquired pneumonia, but not specifically for proton pump inhibitor therapy;41, 43, 44 one study did not perform adjusted analysis for proton pump inhibitor use and the risk of community-acquired pneumonia;42 another study included more than one episode of pneumonia per person (i.e. multiple outcomes per person)21 and yet another study included results of randomized controlled trials.45 The weighted Kappa score for agreement was 0.91 (Figure 1).

Figure 1.

 Summary of search and selection of articles. *See text for details.

Six studies involving approximately 1 million patients were included in this review and are summarized in Table 3.15–20 All were large population-based case-control studies. One was conducted in Canada20 and five were conducted in Europe.15–19 In particular, two of the European studies18, 19 used the same large United Kingdom database. We justified including both studies in the meta-analysis as they were not duplicate publications; they were written by different authors at different institutions, had different patient eligibility criteria and different study periods. Importantly, the ages of the study population in the study by Rodriguez et al. ranged from 20 to 79 years, whereas the study population in the study by Myles et al. included all individuals 40 years and older.18, 19 Specifically, 32.5% of the population in the study by Myles et al. was 80 and over.18 Thus, by excluding the study by Myles et al. outright, much important data may be lost. Similarly, excluding the study by Rodriguez et al. would result in the loss of considerable information as their study spanned from 1 January 2000 to 31 December 2005 whereas the study by Myles only ranged from 1 July 2001 to 1 July 2002.18, 19 By including both studies, we captured the maximum number of participant-events (Table 3). However, given that there was a small amount of overlap between the two databases, we performed a post hoc sensitivity analysis to examine the effect of each of these two studies on the primary outcome and on heterogeneity.

Table 3.   Characteristics of individual studies
StudySample sizeInclusion/ExclusionMethodsPopulationInterventionOutcome measured
  1. CAP, community-acquired pneumonia; UK, United Kingdom; HCAP, healthcare-associated pneumonia.

  2. * Proportion of types of proton pump inhibitors used not specified.

  3. † Dose and duration of proton pump inhibitor not specified.

Laheij 2004N = 5165Inclusion:
Individuals with ≥1 year of database history
Exclusion:
Individuals with <1 year of database history
Individuals with pneumonia in pre-enrolment period
Treatment for H. pylori
Nested case control
Cases: 475
Controls: 4690
Matched 1:10
Age
Gender
Index date of case
Population-based study
January 1995–December 2002 (the Netherlands)
Proton pump inhibitor
All types specified
Doses specified
Duration of use specified
Risk of CAP
Hospitalized/out-patient
Gulmez 2007N = 41818Inclusion:
All patients within County of Funen
Exclusion:
Malignancy
Hospitalization within 7 days of index date
Case control
Cases: 7642
Controls: 34176
Matched 1:4
Age
Gender
Population-based study
January 2000–December 2004
(Denmark)
Proton pump inhibitor*
Doses specified
Duration of use specified
Risk of CAP
Hospitalized cases
Sarkar 2008N = 879947Inclusion:
>18 years
Exclusion:
CAP within 6 months of enrolment
Aspiration pneumonia
Treatment for H. pylori
Nested case control
Cases: 80066
Control: 799881
Matched 1:10
Index date of enrolment
General practice
Population-based study
May 1987–April 2002
(UK)
Proton pump inhibitors*
Doses specified
Duration of use specified
Risk of CAP
Hospitalized/out-patient
Myles 2009N = 25883Inclusion:
Individuals >40 years
Exclusion:
None listed
Nested case control
Cases: 3709
Control: 22174
Matched 1:6
Age
Gender
Practice
Population-based study
(UK)
July 2001–July 2002
Proton pump inhibitor*†Risk of CAP
Out-patient
Rodriguez 2009N = 17290Inclusion:
Age 20–79 years
For cases, only definite or possible cases of CAP after manual review of patients computerized clinical profile
Exclusion:
For cases, subjects with discharge diagnosis of HCAP, if CAP was only listed as cause of death and prevalent cases of CAP
Nested case control
Cases: 7297
Control: 9993
Matched
Age
Gender
Index date
Population-based study (UK)
1 January 2000–31 December 2005
Proton pump inhibitor
Doses specified
Duration of use specified
Risk of CAP
Hospitalized/out-patient
Eurich 2010N = 2724Inclusion:
Individuals ≥65 years
Hospitalization with CAP
Exclusion:
Known TB, CF, immunocompromised or pregnant at time of enrolment
Re-admission within 30 days of discharge from index admission
Nested case control
Cases: 248
Control: 2476
Matched 1:10
Age
Gender
Population-based study (Canada)Proton pump inhibitor*
Duration of use specified
Risk of CAP
Hospitalized cases

All six studies assessed the risk of community-acquired pneumonia as an outcome;15–20 participants were mainly older (≥60 years) with a similar number of men and women (Table 4).

Table 4.   Baseline patient characteristics
AuthorAge ≥60 yearsMaleSmokerCOPDStrokeGERDImpaired functional status
  1. * Adjusted for this variable, but did not report the proportion.

  2. † Reported as mean age (s.d.): 73.5 (17.6).

Laheij61%41%Not reported22%Not reported31%Not reported
Gulmez60%53%Not reported6%6%Not reportedNot reported
SarkarNot reported†52%16%11%5%Not reportedNot reported
Myles73%46%20%12%Adjusted*Not reportedNot reported
Rodriguez55%52%38%9%7%13%Not reported
Eurich100%51%15%38%Not reportedNot reported8%

Effects of proton pump inhibitors on risk of community-acquired pneumonia

Six case-control studies15–20 were included in the evaluation of proton pump inhibitor therapy use and risk of community-acquired pneumonia. The odds of developing community-acquired pneumonia in proton pump inhibitor users was 1.36 (95% CI 1.12–1.65) when compared to non-proton pump inhibitor users (Figure 2). However, there was significant heterogeneity (I2 92%, P < 0.001).

Figure 2.

 Forrest plot evaluating the association between proton pump inhibitor use and risk of community-acquired pneumonia.

With respect to severity of illness as an a priori hypothesis for heterogeneity, four studies evaluated the association between proton pump inhibitor use and risk of community-acquired pneumonia requiring hospitalization (Figure 3) and did not show a statistically significant increased risk [OR 1.23 (95% CI 0.99–1.53)]. In contrast, two studies provided data on associated risk of community-acquired pneumonia not requiring hospitalization and showed an increased risk [OR 1.37 (95% CI 1.08–1.75)]; however, there was no significant interaction between these two point estimates (P = 0.52) and significant heterogeneity remained (Figure 3).

Figure 3.

 Forrest plot evaluating the association between proton pump inhibitor use and risk of community-acquired pneumonia in subgroup analysis.

Five studies assessed the impact of duration of therapy on risk of community-acquired pneumonia by stratifying users of proton pump inhibitor therapy into new users and chronic users (Figure 3). The odds of developing pneumonia among new users of proton pump inhibitors was 1.92 when compared to non-proton pump inhibitor users (95% CI 1.40–2.63). In contrast, there was no statistically significant association between community-acquired pneumonia and chronic users of proton pump inhibitor therapy [OR 1.11 (95% CI 0.90–1.38)]. There was a significant interaction between new users of proton pump inhibitors and chronic proton pump inhibitor users (P < 0.005), although significant heterogeneity remained (Figure 3).

Three studies examined the impact of low-dose proton pump inhibitor therapy on risk of pneumonia. There was an association between users of low-dose proton pump inhibitor therapy and community-acquired pneumonia [OR 1.20 (95% CI 1.02–1.43)]. Four studies investigated the impact of high-dose proton pump inhibitors on risk of pneumonia and there was an association with pneumonia and high dose users [1.36 (95% CI 1.16–1.59)]. Although the heterogeneity improved with this subgroup analysis (Figure 3), there was no significant interaction between dosages of proton pump inhibitor therapy (P = 0.29).

Sensitivity analysis

Although all studies initially obtained pneumonia outcomes from established databases, two studies performed sensitivity analyses, providing the association between proton pump inhibitors and ‘confirmed’ community-acquired pneumonia.15, 16‘Confirmed’ pneumonia was defined as patients with both community-acquired pneumonia (from the database coding) and a respiratory pathogen isolated.15, 16 When only ‘confirmed’ cases of pneumonia were evaluated, there was no statistically significant increased risk of community-acquired pneumonia in proton pump inhibitor users [OR 1.52 (95% CI 0.77–2.99)] and significant heterogeneity remained (Figure 4). As no study adjusted for all outlined confounding variables, a sensitivity analysis on this basis could not be performed.

Figure 4.

 Forrest plot estimating risk of ‘confirmed’ community-acquired pneumonia associated with proton pump inhibitor use.

A post hoc sensitivity analysis was also performed, evaluating the impact of each of the two studies from the same United Kingdom database on the primary outcome.18, 19 Removal of the study by Myles et al. only minimally affected the primary outcome [OR 1.32 (95% CI 1.08–1.61)] and had little impact on heterogeneity (I2 91%, P < 0.001). Similarly, removal of the study by Rodriguez et al. did not significantly alter the summary statistic [OR 1.41 (95% CI 1.10–1.83)] or heterogeneity [I2 94%, (P < 0.001)].

Discussion

We found an association between community-acquired pneumonia and proton pump inhibitor therapy use in our review and meta-analysis of six studies that included approximately 1 million people; however, significant heterogeneity limited interpretation of the summary odds ratio. In exploratory analysis, subgroup analysis revealed that risk of community-acquired pneumonia differed depending on duration of use of proton pump inhibitor therapy. In fact, new users of proton pump inhibitor therapy had approximately two times the risk of developing pneumonia when compared with non-users, whereas chronic use was not associated with an increased risk of pneumonia. However, stratification by duration of proton pump inhibitor use did not fully explain the heterogeneity.

The risk of community-acquired pneumonia from proton pump inhibitor therapy observed in these observational studies could be explained by bias and confounding. At present, there is no gold standard for assessment of risk of bias in observational studies.46, 47 However, potential sources of confounding and selection and measurement bias have been identified as important domains to consider when evaluating risk of bias in observational studies.46 Accordingly, risk of bias due to confounding was present in all studies as four did not specifically adjust for GERD,16–18, 20 two did not adjust for history of stroke15, 20 and only one adjusted for functional status.20 Although GERD is not an established risk factor for community-acquired pneumonia, GERD is known to be associated with aspiration pneumonitis, which could be misclassified as pneumonia.40

Measurement bias was also a concern as none of the studies used prospectively collected clinical definitions of pneumonia. Thus, in a sensitivity analysis we evaluated the association between proton pump inhibitor use and risk of ‘confirmed’ cases of pneumonia. Although the summary estimate of 1.52 in the sensitivity analysis was similar to the overall odds ratio of 1.36, the confidence interval of the sensitivity analysis crossed one [OR 1.52 (95% CI 0.77–2.99)]. It remains unknown whether this reflects bias in the overall summary estimate or loss of power in the sensitivity analysis, because of fewer cases analysed. The difference in risk of pneumonia based on duration of proton pump inhibitor use and the consistency of this finding across studies is compelling. However, in addition to the risk of bias described above, some researchers have suggested that these results may be as a consequence of protopathic bias,17 that is, a bias that may arise when drugs are given to treat symptoms (i.e. proton pump inhibitors prescribed for chest pain) and appear temporally associated with a subsequent diagnosis (i.e. community-acquired pneumonia).48

Very few published proton pump inhibitor therapeutic trials have included community-acquired pneumonia as an adverse event22 and to our knowledge, none has stratified the risk of pneumonia by duration of therapy. A retrospective analysis of the AstraZeneca ARIADNE safety database containing the reports of all adverse events from phase II–IV randomized controlled trials, evaluated the risk of community-acquired pneumonia from esomeprazole compared with placebo and found the overall relative risk of pneumonia to be 0.94 (99% CI 0.29–3.07).23 However, when the risk of community-acquired pneumonia was divided into risk from short-term use of esomeprazole vs. risk from longer-term use of esomeprazole, the summary point estimates were quite different. The relative risk of pneumonia with a short 4–8 weeks treatment course was 2.98 (99% CI 0.18–48) when compared with placebo, whereas the relative risk for long-term use of esomeprazole (16–26 weeks of treatment) was 0.64 (99% CI 0.16–2.48) when compared with placebo. Although the confidence intervals of both risk estimates are wide and cross 1 because of the small number of events, the difference between the point estimates is noteworthy.

Thus, although bias and confounding could explain the association between proton pump inhibitor therapy and community-acquired pneumonia in this meta-analysis, the similarity of our risk estimate of pneumonia when duration of therapy was considered with the safety data from randomized controlled trials merits consideration of the possibility that the association could be real.23 Recent experience with newly identified adverse effects of medications has reinforced the precautionary principle and the importance of investigating potential drug-related harms.49

It is hypothesized that proton pump inhibitors may predispose individuals to pneumonia as acid suppressive therapy can lead to overgrowth of bacteria in the stomach50 and subsequently bacterial overgrowth increases the risk for micro-aspiration of bacteria.51 However, the mechanism by which differing durations of proton pump inhibitor therapy use could impact risk of pneumonia remains unknown. Proton pump inhibitors are able to inhibit cellular components of the innate and adaptive immune system including the activity of neutrophils, natural killer cells and cytotoxic T-cell activity.52–54 We speculate that with time, the immune system could develop ways to compensate or overcome these immune deficits.

Conclusions

In this meta-analysis of six observational studies, proton pump inhibitor therapy was associated with community-acquired pneumonia; however, significant heterogeneity limits the interpretation of the summary odds ratio. In exploratory analysis, the risk of community-acquired pneumonia was two times higher in newly prescribed proton pump inhibitor users than in non-users, whereas there was no difference in risk of pneumonia in chronic users of proton pump inhibitors. These results could be explained by bias and confounding; however, the possibility remains that proton pump inhibitors could be associated with an increased risk of community-acquired pneumonia, particularly within 30-days. Based on the findings of this review, future randomized controlled trials of proton pump inhibitors should include community-acquired pneumonia as a potential adverse effect. In the interim, observational studies should be conducted to help clarify the association between new users of proton pump inhibitors and the risk of community-acquired pneumonia.

Acknowledgements

Declaration of personal interests: None. We thank Jason Busse for his input and advice and Neera Bhatnagar for her helpful literature search suggestions. We also thank Dr T. Marrie for his help as a content expert. Declaration of funding interests: Dr Johnstone receives salary support from the Bayer Healthcare Research Fellowship at McMaster University and Dr Nerenberg received a salary support award from Canadian Institutes of Health Research.

Ancillary