Risk of Community-Acquired Pneumonia with Outpatient Proton-Pump Inhibitor Therapy: A Systematic Review and Meta-Analysis

1Community-acquired pneumonia (CAP) is a common diagnosis associated with substantial morbidity and healthcare expenditure. In 2006 alone, 4.2 million ambulatory care visits for CAP occurred in the United States. Medicare data from 2007–2008 indicated a 30-day mortality ranging from 3.8 to 8.5% depending on severity of disease. Annual healthcare costs incurred by patients with CAP are estimated to be approximately $13 billion among Medicare fee-for service patients. Implementation of guidelines for antibiotic selection and administration of pneumococcal vaccination have been shown to reduce CAP incidence, morbidity and mortality. Identification and avoidance of medications associated with an increased risk of CAP could further reduce CAP incidence.

2Proton pump inhibitors (PPIs) are among the most widely prescribed medications. In 2011, omeprazole was the sixth most commonly prescribed medication in the United States with nearly 60 million prescriptions. PPIs have become a 10 billion dollar industry with over 15 million Americans taking these medications, not including over-the-counter usage. Although evidence and guidelines support the use of PPIs for gastroesophageal reflux disease (GERD) and select cases of duodenal and gastric ulcers, evaluation of PPI therapy in the ambulatory setting suggests that as few as 35% of patients taking PPIs have an appropriate indication documented. A spectrum of side effects are associated with PPI therapy, including deficiencies of critical vitamins and minerals, Clostridium difficile-associated diarrhea infection and hip fracture. CAP has been hypothesized as an additional untoward consequence of PPI therapy.

3We conducted a systematic review and meta-analysis of observational and randomized studies to determine whether outpatient PPI therapy is associated with increased risk of CAP among adults as compared to no PPI therapy. To identify particularly high-risk patients, we conducted additional analyses to determine the CAP risk stratified by PPI dose, duration of PPI therapy, and age of participant.

4Our systematic review and meta-analysis was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) and Meta-analysis of Observational Studies in Epidemiology (MOOSE) (PRISMA checklist reported in S1 Table). A written review protocol was not drafted.

5We performed systematic searches of MEDLINE (via PubMed), EMBASE, CINAHL, Cochrane Central Register of Controlled Trials (CENTRAL), Scopus, Web of Science and ClinicalTrials.gov on February 3, 2014. Our search strings included controlled vocabulary and related keywords for two concepts: pneumonia and acid suppressants (S2 Table). We did not limit our search based on study design, publication year, language or inclusion of human participants. Authors of potentially eligible abstracts, posters or manuscripts were contacted via email to clarify study information and obtain additional data. Approval for use of unpublished data in our analysis and certification of data validity was documented electronically.

6We included in our review studies reporting data to allow the calculation of a measure of risk of CAP among adult participants with and without exposure to outpatient PPI therapy. We followed guidance from the Cochrane Collaboration to include both observational and randomized studies to maximize study, participant and event inclusion into our analysis and to thereby allow the conduct of sensitivity and subgroup analyses. We excluded case reports and case series. We screened systematic reviews for the inclusion of data not published elsewhere and for references not otherwise captured by our search strategy. We excluded studies in which <95% of participants were over 18 years old. We also excluded studies if PPIs were exclusively administered intravenously, in the inpatient setting, peri-procedurally, or as part of Helicobacter pylori therapy. CAP cases were identified by the definitions utilized in each included study. We excluded studies in which CAP preceded PPI exposure or in which the temporal relationship was ambiguous.

7Two authors independently screened studies for inclusion and a third author adjudicated discordant assessments. Title/abstract and full text screening were conducted in a similar fashion; however, specific exclusion reasons were documented only during full text screening. Upon selection of the final group of studies, two authors independently extracted qualitative and quantitative data using a standardized data extraction form adjudicated by a third author. To assess the methodological quality of observational studies, we used a modified version of the Newcastle-Ottawa Scale (S3 Table). We applied this validated tool to characterize participant selection, comparability of populations, and outcome assessment.

8The primary outcome of this meta-analysis was incident CAP during treatment with outpatient PPI therapy. Sensitivity analyses examined our primary outcome among studies in which PPI therapy was the single form of gastric acid suppression, studies with our strict definition of CAP that included radiographic confirmation, and studies with lower risk of bias (defined as low risk on ≥4 out of 7 criteria for cohort studies and ≥6 out of 8 criteria for case-control studies using the modified Newcastle-Ottawa Scale). Secondary analyses evaluated the risk of CAP with H2-receptor antagonist (H2RA) therapy and risk of hospitalization for CAP with PPI therapy. Lastly, we examined three subgroups of participant populations in order to more specifically characterize the risk of CAP associated with PPI therapy: those treated with different PPI doses (high dose > 1 defined daily dose [DDD] and low dose PPI ≤ 1 DDD), durations of PPI therapy prior to CAP diagnosis (<1 month, 1–6 months, >6 months), and age categories (<65 years old or ≥65 years old). All subgroup analyses were defined a priori.

9We report measures of association extracted from the included studies: odds ratios (ORs), relative risks (RRs) and hazard ratios (HRs), each with a 95% confidence interval (CI). For studies that did not report a measure of association between PPI exposure and CAP but did include the required information, we calculated an unadjusted odds ratio using the Peto method.

10We generated a funnel plot to visually assess reporting bias and used Egger’s test to assess asymmetry of the funnel plot. In our qualitative analysis, we critically examined the study population, study design, internal and external validity, and exposure and outcome ascertainment and definitions.

11To determine the proportion of variability in the effect estimates due to heterogeneity, we calculated the I2 statistic. We adhered to the accepted cutoff of 50% to define significant heterogeneity and additionally calculated the associated p-value (via χ2 test). We generated forest plots for each of our exposure-outcome comparisons of interest. Because of significant clinical, methodological and statistical heterogeneity across the included studies, we used a random-effects based model to calculate the weighted mean, variance of the summary effect, and associated 95% confidence interval and p-value. Since pneumonia is a rare clinical event, all effect estimates were assumed to approximate the relative risk and were pooled in meta-analyses. All statistical analyses were conducted using STATA v12.

12Our systematic search identified 5789 unique titles and abstracts. Of these, 5410 studies were excluded based on predetermined eligibility criteria during title/abstract review. The remaining 379 studies underwent full text screening and 33 ultimately met criteria for inclusion in our systematic review. Due to overlapping study populations, 26 of these 33 studies were included in our meta-analysis. Of the 346 excluded full texts, the most common reason for exclusion was a lack of CAP assessment (n = 131), followed by absence of an appropriate PPI exposure group (n = 62), and ineligible study design (n = 53). Reasons for exclusion are detailed in the PRISMA flow diagram (Fig 1).

13A qualitative synthesis of the 33 studies meeting our inclusion criteria is summarized in Table 1 and S4 Table. The majority of studies were of case-control design (n = 18); 10 were cohort studies, 4 randomized controlled trials (RCTs), and 1 case-crossover study. Studies were conducted across the world with the United States and the United Kingdom contributing data to the most studies (11 and 10 studies, respectively). Data collection spanned over 2 decades (1987–2011).

14For the majority of studies, PPI exposure could include any available drug in the class. However, two RCTs administered esomeprazole only and two RCTs administered lansoprazole only. Four studies included exposed participants who were treated with either PPI or H2-receptor antagonist therapy (H2RA). Most studies used pharmacy or prescription records to identify PPI exposure (n = 24, 73%) and diagnostic codes to identify CAP cases (n = 23, 70%).

15A total of 6,546,396 study participants were included in our qualitative synthesis. Studies included participants with a breadth of comorbidities which are described in S4 Table; those studies examining CAP within specific disease populations are noted in Table 1. Examples include studies restricting to persons taking anti-Parkinsonian medications, to those with newly initiated NSAID therapy, to persons with a stroke history, or to persons with prior kidney transplantation. total of 262,906 cases of CAP were reported. One study identified CAP cases managed exclusively in the outpatient setting; all other studies presented inpatient or a mix of inpatient and outpatient CAP cases. Individual study definitions of CAP are outlined in S4 Table.

16The included observational studies generally demonstrated good methodologic quality, as assessed by a modified Newcastle-Ottawa Scale (S1 Fig). Case-control studies appropriately conducted and clearly reported exposure and outcome ascertainment. Only one study documented the non-response rate. Cohort studies demonstrated greater risk of bias than case-control studies, particularly with regards to assessment of CAP and follow-up of participants. Among the cohort studies, three showed particularly high risk for bias with respect to our association of interest. The Cochrane Collaboration’s tool was used to assess sequence generation, allocation concealment, and blinding in the included RCTs; all satisfied the requirements for a low risk of bias in these areas. The tool’s assessments of incomplete outcome reporting and selective outcome reporting, however, are not relevant to the studies included in this review because none of these included CAP as a primary outcome. To address the potentially high risk of bias associated with RCTs not designed to assess the association between PPI exposure and CAP, a sensitivity analysis was conducted excluding RCTs from the meta-analysis.

17Our funnel plot (S2 Fig) showed clustering of studies below our summary effect estimate. Egger’s test confirmed asymmetry of the funnel plot (0.27; 95% CI 0.14, 0.40; p<0.001). This may reflect a publication bias attenuating our results.

18Our meta-analysis pools data from 26 of the 33 studies included in the systematic review. These 26 studies include 6,351,656 (97.0%) participants and 226,769 (86.2%) cases of CAP. Seven studies were not included in the meta-analysis due to substantially overlapping participant populations.

19Of the 26 studies included in our primary meta-analysis, 15 reported a statistically significant increased risk of CAP with PPI use and 11 reported no statistically significant association (Fig 2; S3 Fig). The 15 studies demonstrating significantly increased risk included 138,593 of the total reported 226,769 cases of CAP (61.1%). Four studies reported unadjusted ORs. The Peto method was used to calculate an unadjusted OR for 7 studies. Collectively, our meta-analysis identified a significantly increased risk of CAP with outpatient PPI use, with a pooled RR of 1.49 (95% CI 1.16, 1.92). The I2 statistic for this summary measure was 99.2% (p<0.001), indicating significant heterogeneity across studies.

20When limiting our analysis to studies that measured PPI exposure separately from other gastric acid suppressant drugs, we found that the association between PPI exposure and CAP remained similar to our primary analysis and heterogeneity remained high (RR 1.48; 95% CI 1.14, 1.92; I2 99.2%). When limiting our analysis to the 6 studies meeting our strict CAP definition which included radiographic confirmation, the increased risk of CAP with PPI therapy was attenuated and heterogeneity was reduced (RR 1.22; 95% CI 0.99, 1.52; I 2 65.8%). When we excluded RCTs, point estimates and confidence intervals were nearly identical to those from the primary analysis (RR 1.49; 95% CI 1.15, 1.93; I2 99.3%). Our results were also robust to sensitivity analyses restricting to the 13 studies identified as having lower risk of bias and to studies reporting adjusted measures of association (S5 Table).

21We pooled 16 studies in order to evaluate the secondary outcome of hospital admission for CAP among PPI users compared to non-PPI users. The pooled relative risk of hospitalization with CAP was 1.61-fold higher among PPI users as compared to non-PPI users (95% CI: 1.12, 2.31; I2 99.3%).

22In order to evaluate CAP risk with H2RA therapy, we pooled 8 studies that reported this risk separately from the risk associated with PPI therapy. Combined, these studies included 4837 cases of CAP. When pooled, these studies demonstrated no significant CAP risk with H2RA therapy as compared to no acid-suppression therapy (RR 1.00; 95% CI 0.90, 1.12; I2 33.7%).

23Subgroup analyses are summarized in Table 2. The 8 studies reporting risk of CAP with low dose PPI therapy (≤ 1 DDD), as compared to no PPI therapy, identified a pooled RR of 1.31 (95% CI 1.04, 1.66), which was similar to findings among those taking high dose PPI therapy (RR 1.33; 95% CI 1.05, 1.69). Treatment with a PPI for less than 1 month was associated with the highest risk of CAP (OR 2.10; 95% CI 1.39, 3.16), as compared to no PPI therapy. The magnitude of CAP risk decreased and lost statistical significance as duration of PPI therapy increased (OR 1.51; 95% CI 0.92, 2.49 for 1–6 months and OR 1.37; 95% CI 0.85, 2.20 for >6months). Participants aged ≥ 65 years demonstrated significantly increased risk of CAP with PPI therapy as compared to those not taking PPIs (OR 1.33; 95% CI 1.13, 1.58; I2 85.4%). Similar findings were observed among participants <65 years old (OR 1.34; 95% CI 1.04, 1.71; I2 60.1%).

24This systematic review of 33 studies and meta-analysis of 26 studies demonstrated a 1.5-fold increased risk of community-acquired pneumonia with outpatient proton-pump inhibitor therapy. PPI therapy was also associated with an increased risk for hospitalization with CAP (1.6-fold). No association was observed between H2RA use and CAP among studies examining participants taking H2RA therapy alone. CAP risk did not vary by PPI dose or participant age but more than doubled among those treated for less than 1 month, as compared to those not taking PPIs. Our findings were robust to sensitivity analyses restricting to studies exploring only PPI therapy (excluding H2RA therapy), studies utilizing a CAP definition requiring radiographic confirmation, and studies found to have lower risk of bias during qualitative analysis.

25Several pathogenic mechanisms have been proposed to explain the association between PPI use and incidence of CAP. Decreased gastric acidity is associated with alteration of gut flora. Micro-aspiration of the altered gut flora is one hypothesized mechanism for the increased CAP risk observed in the setting of elevated gut pH. More recently, proton pumps have been localized to the upper and lower respiratory tract, suggesting that pH dysregulation may additionally alter respiratory flora, thereby directly inducing infection. Our findings are consistent with the theory that PPI therapy may lead to CAP both through acute pH dysregulation and alteration of the gut microbiome. CAP risk was greatest during the first month of therapy, which is the time period during which the aero-digestive microbiome may be in greatest flux. This association between CAP risk and short duration of PPI therapy, in conjunction with a lack of dose and duration effect, suggest that the acute changes occurring with onset of PPI therapy may be responsible for CAP risk. Our findings cannot define a causal relationship but do highlight an important association that requires further investigation. CAP risk was absent among participants taking H2RAs, which have weaker acid suppressant activity than do PPIs, however our study was not constructed to fully investigate this question.

26Given the widespread usage of PPI therapy, often without an appropriate indication, the excess risk of CAP among PPI users could translate into a substantial burden on the healthcare system. Moreover, the increased risk of hospitalization for CAP underscores the potential clinical and financial impact of this adverse effect. Careful consideration of the risks, benefits and alternative treatment options should occur with all PPI prescriptions. Our observation that CAP risk was increased with PPI therapy, regardless of PPI dose or participant age, implies that alternate therapies, when appropriate, may be a strategy for reducing CAP risk. Secondary analysis showing an absence of CAP risk among participants taking H2RAs further supports the reduced risk profile of alternate therapies.

27Our systematic review and meta-analysis could be subject to several sources of bias. Confounding by indication may have augmented our findings if the incidence of CAP is higher among persons taking PPIs due to the nature of their disease state, unrelated to the PPI therapy. GERD, a common indication for PPI therapy, may directly increase risk of CAP by increasing microaspiration of gut flora. Several studies included in our review utilized different methods to reduce this source of confounding. Four studies restricted study enrollment to participants with a non-GERD indication for PPI, one study employed a case-crossover design so that participants in both the exposed and unexposed groups possessed the same comorbidities, and two studies compared current PPI users to former PPI users to limit variability in comorbidities between cases and controls. Jena, et al., further explored the concept of confounding by using prescription drug and medical claims data to demonstrate statistically significant associations between PPI use and conditions without a readily apparent pathophysiologic link to PPI use, such as osteoarthritis, chest pain, and urinary tract infection. The authors raise the important point that the observed relationship between PPI use and CAP may be influenced by unmeasured confounders. This warrants further investigation, including replication of the findings using clinically-driven criteria for disease outcomes and efforts to elucidate specific confounders that may be driving any associations. Importantly, the absence of a known pathophysiologic mechanism for a causal relationship does not preclude the existence of one.

28Reporting bias is an additional source of bias inherent to systematic reviews and meta-analyses. Our search was designed to detect a broad array of study designs respiratory diagnoses, yet only identified four RCTs reporting pneumonia as an adverse event. The four RCTs included in our review contribute few patients to the primary analysis, were not designed specifically to address our study question, and relied upon adverse event reporting to assess the incidence of pneumonia. Still, our findings were essentially unchanged in multiple sensitivity analyses, including an analysis that excluded RCTs.

29Publishing bias may have also influenced our findings. Though our funnel plot was limited by the number of included studies, it suggested a publication bias toward a protective or null effect of PPI therapy. If present, the magnitude of the association we report may actually be smaller than the true effect.

30Our study has limitations. Several studies reported data on overlapping participant populations. We selected only the most inclusive studies but cannot guarantee inclusion of all eligible participants in our meta-analysis. The included studies reported one of three different measures of association: odds ratio, hazard ratio, or relative risk. Because CAP is a rare outcome, the odds ratio and hazard ratio are each assumed to approximate the relative risk. We observed the greatest risk for CAP in the setting of PPI use for less than 1 month, which may have a biological explanation but may also, at least partly, reflect a protopathic bias due to initiation of PPI therapy around the time of the development of respiratory symptoms.

31Tests for heterogeneity demonstrated variability of findings across studies; however consistent demonstration of increased CAP risk with PPI therapy and the findings from our qualitative analysis suggested that pooling of the 26 identified studies for meta-analysis was reasonable despite differences in methodology. The impact of these differences was assessed in subgroup, secondary, and sensitivity analyses. In many cases, studies demonstrated significantly less heterogeneity when grouped for these additional analyses. The subgroup analysis of the association between high-dose PPI exposure and CAP demonstrated the least heterogeneity (I2 34.0%). Almost all of these analyses supported our primary findings. Among sensitivity analyses, only the analysis examining the strictest CAP definition failed to show a significant association between PPI exposure and CAP. With only 7 included studies, representing 3,865 CAP cases among 16,537 participants, this sensitivity analysis was likely underpowered to detect a significant difference.

32Several previous meta-analyses have sought to evaluate the relationship between PPI therapy and CAP. Giuliano, et al, concluded that patients prescribed PPIs had an increased risk for CAP, particularly at high PPI doses and soon after initiation of PPI therapy. Eom, et al., found that patients prescribed any acid suppressant therapy were at increased risk of community-acquired or hospital-acquired pneumonia. Our meta-analysis enhances the existing literature by adding data from 19 studies that were not included in these prior investigations. These 19 studies were identified through a combination of our broad, updated database search and our extensive pursuit of unpublished data. The inclusion of unpublished data also strengthened our subgroup and sensitivity analyses.

33In conclusion, we conducted a thorough systematic review and rigorous meta-analysis of studies reporting the risk of CAP with ambulatory PPI therapy. We identified a 1.5-fold increased risk for CAP with outpatient PPI therapy, with additional analyses revealing even greater risk during the first month of therapy and for CAP hospitalization. Our findings add to currently available literature through the breadth of our search, inclusion of unpublished data and updated research, and extent of analysis. Given the morbidity and mortality associated with CAP and the extent of PPI use in the United States, identification of any risk associated with PPI use is critical for risk stratification and modification where possible. Future studies should employ a rigorous definition of CAP and ascertain PPI exposure at various doses and durations to further explore the association between PPI exposure and CAP within specific populations. Studies are also needed to clarify any pathophysiologic mechanisms underlying the observed association. We recommend careful consideration of the risks and benefits when initiating PPI therapy and heightened awareness regarding this risk factor for adults presenting with pneumonia.