Introduction
Cirrhosis accounts for approximately 1 million deaths annually [1]. Common complications of liver cirrhosis include variceal bleeding, ascites, hepatic encephalopathy, hepatorenal syndrome, and spontaneous bacterial peritonitis (SBP) [2]. In addition, cirrhosis is associated with an increased risk and poorer prognosis of several other types of bacterial infection, such as bacteremia, enterocolitis, skin and soft tissue infection, urinary tract infection, meningitis, and infective endocarditis [3]. Cirrhosis-associated immune dysfunction is thought to be the key player in this increased risk [4].
Proton pump inhibitors (PPI) are commonly prescribed for cirrhotic patients, although studies have shown that over half of PPI use was without valid indications [5], which may lead to an unnecessary increased incidence of several complications, such as hepatic encephalopathy, Clostridium difficile infection and SBP [6]. The possible explanation for the increased likelihood of developing SBP among PPI users is that acid suppression facilitates bacterial overgrowth and translocation [7-9]. The use of PPI may also lead to an increased risk of other types of organ-specific bacterial infection. In fact, studies have suggested an increased risk of bacterial pneumonia among cirrhotic patients who use PPIs, although the results are inconsistent [10-17]. The current study aimed to further investigate this risk by identifying all available studies and summarizing their results together.
Materials and methods
Information sources and search strategy
A systematic literature review based on the EMBASE and MEDLINE databases was performed independently by 2 investigators (WW and NC) from inception to September 2019 to identify all published studies that examined the risk or association between pneumonia and PPI use in cirrhotic patients. The search strategy, which included the terms “proton pump inhibitors” and “cirrhosis”, is available as Supplementary Table 1. In addition, we manually reviewed the references of the eligible studies to identify any additional potential articles. This study was performed according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses statement (Supplementary Table 2).
Selection criteria
To be eligible, a study had to be an observational study (cohort, case-control or cross-sectional study) that included one group of cirrhotic patients with PPI use and another group of cirrhotic patients without PPI use.
-
Eligible cohort studies started with recruitment of cirrhotic patients who used and did not use PPIs and followed them for incident pneumonia. Relative risk (RR), incidence rate ratio (IRR), hazard risk ratio (HR) or standardized incidence ratio (SIR) with associated 95% confidence interval (CI) comparing the incidence of pneumonia between cirrhotic patients with and without PPI use had to be provided.
-
Eligible case-control studies started with recruitment of cases of cirrhotic patients with pneumonia and controls who were cirrhotic patients without pneumonia and explored their history of PPI use. Odds ratio (OR) with associated 95%CI comparing the prevalence of PPI use between cases versus controls had to be reported.
-
Eligible cross-sectional studies recruited cirrhotic patients and explored the history of PPI use and pneumonia at the same time. OR with associated 95%CI of this association had to be reported. No language limitation was applied during the systematic review.
Data extraction
We used a standardized data collection form to extract the following information: last name of the first author, country where the study was conducted, study design, year of publication, number of participants, recruitment of participants, how the diagnosis of pneumonia and ascertainment of PPI use were justified, follow-up period and duration (for cohort studies), baseline characteristics of participants, confounders adjusted in multivariate analysis and adjusted effect estimates with corresponding 95%CI. We appraised the quality of the included cohort and case-control studies using the Newcastle-Ottawa quality assessment scale [18]. The modified version of this scale was used for cross-sectional studies.
Statistical analysis
We utilized Review Manager 5.3 software from the Cochrane Collaboration (London, United Kingdom) to analyze all data. Point estimates and standard errors from each study were pooled together using the generic inverse variance method of DerSimonian and Laird, which assigns the weight of the study in reverse to its variance [19]. A random-effect model, rather than a fixed-effect model, was used, as the assumption of the fixed-effect model that every study should give rise to the same result is not justified under almost all circumstances, especially in a meta-analysis of observational studies. Statistical heterogeneity was assessed by Cochran’s Q test, complimented by the I2 statistic. This I2 statistic quantifies the proportion of total variation across studies due to heterogeneity rather than chance. A value of I2 of 0-25% represents insignificant heterogeneity, 26-50% represents low heterogeneity, 51-75% represents moderate heterogeneity, and 76% or higher represents high heterogeneity [20]. The presence of publication bias was assessed by visualization of a funnel plot along with Egger’s regression test. Egger’s regression test was conducted using Comprehensive Meta-analysis 3.0 software (Englewood, New Jersey, United States).
Results
A total of 1947 articles (276 from MEDLINE and 1671 from EMBASE) were identified, from which 237 duplicated articles were removed, leaving 1710 articles for title and abstract review. At this stage of review, 1675 articles were excluded because they were clearly ineligible based on study design and type of article. Therefore, 35 full-length articles were thoroughly reviewed, and 27 articles were further excluded as they did not report the outcome of interest. Finally, 3 cohort and 5 cross-sectional studies with a total of 40,295 participants met the eligibility criteria and were included in the meta-analysis [10-17]. Two of the 3 included cohort studies were published as conference abstracts [10,14]. No eligible case-control study was identified. Fig. 1 summarizes the literature review and study selection process. Tables 1 and 2 describe the characteristics and Newcastle-Ottawa assessment scales of the included cross-sectional and cohort studies, respectively.
Figure 1 Literature review and study selection process
Table 1 Main characteristics of the cross-sectional studies included in this meta-analysis