Proton pump inhibitor use and mortality in patients with cirrhosis: a meta-analysis of cohort studies

Abstract Background: Proton pump inhibitor (PPI) is commonly used in patients with cirrhosis. However, some studies demonstrated that PPI use was associated with adverse outcome in patients with cirrhosis. We aimed to perform a meta-analysis of cohort studies to evaluate the association between PPI use and mortality in cirrhotic patients. Methods: Relevant studies were obtained via search of PubMed and Embase databases. A randomized-effect model was used to pool the results. Subgroup analyses were performed to evaluate the source of heterogeneity. Results: Overall, 21 cohort studies with 20,899 patients and 7457 death events were included. The pooled results with a randomized-effect model showed that PPI use was associated with significantly increased risk of mortality in patients with cirrhosis (adjusted relative risk [RR] = RR: 1.39, P<0.001) with considerable heterogeneity (I2=73%). Subgroup analyses showed that characteristics such as patient ethnicity, sample size, definition of PPI use, and complications of patients did not affect the association. However, the association between PPI use and mortality was independent of study characteristics including patient ethnicity, sample size, complications, definition of PPI use, and follow-up duration. However, the association between PPI use and mortality in cirrhotic patients was significant in retrospective studies (RR: 1.40, P<0.001), but not in prospective studies (RR: 1.34, P=0.33). Conclusions: PPI use may be associated with moderately increased mortality in cirrhotic patients. Although prospective cohort studies are needed to validate our findings, PPI should only prescribed to cirrhotic patients with indications for the treatment.


Introduction
Proton pump inhibitors (PPIs) are the first-line medications for the treatment of acid-related diseases, such as gastroesophageal reflux disease, peptic ulcer, and Zollinger-Ellison syndrome et al [1][2][3]. For patients with cirrhosis, PPIs are also commonly used. According to previous data, proportions of cirrhotic patients who were prescribed with PPIs varied from 30% to 80% [4]. However, the indications and clinical efficacy of PPIs in these patients remain to be validated, which highlights the possible overutilization of PPIs in patients with cirrhosis [5]. Subsequently, increasing evidence from epidemiological studies indicated that PPI use may be associated with adverse outcomes in cirrhotic patients, spontaneous bacterial peritonitis (SBP) [6][7][8][9][10][11], and hepatic encephalopathy (HE) [12,13]. However, it remains unknown whether PPI use affects mortality in these patients. Previous studies evaluating the association between PPI use and mortality risk in patients with cirrhosis showed inconsistent results . An early meta-analysis published in 2015 showed that PPI use was not associated with increased mortality in patients with cirrhosis [11]. However, only four cohort studies [14,16,17,19] were included in the present study, and many related cohort studies have been published since the previous meta-analysis [18,[20][21][22][23][24][25][26][27][28][29][30][31][32][33][34]. In addition, it remains unknown whether study design such as study design, ethnicity of the patients, definition of PPI use, or follow-up duration may affect the outcome. Therefore, we aimed to perform a meta-analysis to systematically evaluate the potential association between PPI use and mortality in patients with cirrhosis.

Methods
The meta-analysis was performed in accordance with the MOOSE (Meta-analysis of Observational Studies in Epidemiology) [35] and Cochrane's Handbook [36] guidelines.

Literature search
Studies were identified via systematic search of electronic databases of PubMed and Embase via the following terms: (1) "proton pump inhibitor" OR "proton pump inhibitors" OR "acid suppressive therapy" OR "anti-secretory therapy" OR "PPI" OR "anti-ulcer agent" OR "antacid" OR "omeprazole" OR "esomeprazole" OR "lansoprazole" OR "pantoprazole" OR "rabeprazole" OR "ilaprazole" and (2) "cirrhosis" OR "cirrhotic" OR "liver fibrosis". This extensive search strategy was used to avoid the potential missing of related studies. The search was limited to human studies in English. The reference lists of related original and review articles were also analyzed using a manual approach. The final literature search was performed on September 20, 2019.

Study selection
The inclusion criteria for the studies were: (1) cohort studies published in full-length articles in English; (2) included patients with cirrhosis; (3) evaluated the association between PPI use and mortality risk in these patients; and (4) reported the relative risk for the association after adjustment of potential confounding factors. Reviews, editorials, preclinical studies, and studies irrelevant to the aim of current meta-analysis were excluded.

Data extracting and quality evaluation
Literature search, data extraction, and quality assessment of the included studies were performed according to the predefined inclusion criteria by two independent authors (X.W. and X.L.). If discrepancies occurred, they were resolved by consensus of the two authors. The extracted data included: (1) name of first author, publication year and country where the study was performed; (2) study design characteristics; (3) ethnicity, characteristics, age, and gender of the participants; (4) definition of PPI use; (5) follow-up durations for cohort studies; and (6) variables that were adjusted when presenting the results. The quality of each study was evaluated using the Newcastle-Ottawa Scale [37] that ranges from 1 to 9 stars and judges each study regarding three aspects: selection of the study groups, the comparability of the groups, and the ascertainment of the outcome of interest.

Statistical analyses
We used risk ratios (RRs) and their corresponding 95% confidence intervals (CIs) as the general measure for the association between PPI use and mortality in patients with cirrhosis. Data of RRs and their corresponding stand errors (SEs) were calculated from 95% CIs or P values, and were logarithmically transformed to stabilize variance and normalized the distribution [36]. The Cochrane's Q test and I 2 test were used to evaluate the heterogeneity among the include cohort studies [38]. A significant heterogeneity was considered if I 2 >50%. We used a randomized-effect model to synthesize the RR data because this model is considered as a more generalized method that incorporates of the potential heterogeneity [36]. Sensitivity analyses, by removing individual study one at a time, were performed to test the robustness of the results [39]. Predefined subgroup analyses were performed to evaluate the influences of study characteristics on the outcome, including study design, patient ethnicity, patient characteristics, sample size, definition of PPI use, and follow-up durations. The potential publication bias was assessed by funnel plots with the Egger regression asymmetry test [40]. We used the RevMan (Version 5.1; Cochrane Collaboration, Oxford, U.K.). and STATA software for the meta-analysis and statistics.

Literature search
The process of database search is summarized in Figure 1. Briefly, 1312 articles were found via initial literature search of the PubMed and Embase databases, and 1264 were excluded through screening of the titles and abstracts mainly because they were not relevant to the purpose of the meta-analysis. Subsequently, 48 potential relevant records underwent full-text review. Of these, 27 were further excluded because two of them did not include patients with cirrhosis, one did not analyze PPI in cirrhotic patients, 15 did not report mortality outcome, two did not provide adjusted data for the association, and the other seven were abstracts of the already included studies. Finally, 21 cohort studies were included .

Association between PPI use and mortality in patients with cirrhosis
Meta-analysis with a randomized-effect model including 21 studies showed that use of PPI was significantly associated with mortality in cirrhotic patients  Table 4). However, the difference between the subgroups was not significant (P=0.88). The association between PPI use and mortality in cirrhotic patients was significant and not affected by study characteristics including patient ethnicity, with or without severe complications, sample size, definition of PPI use, or follow-up durations (Table 4).

Publication bias
The funnel plots regarding the association between PPI use and mortality in cirrhotic patients were shown in Figure  3. The funnel plots were symmetry on visual inspection, suggesting low chance of significant publication bias. Results of Egger's regression test also suggested that no significant publication bias (P=0.65). Abbreviations: CI, confidence interval; RR, risk ratio. Abbreviations: CI, confidence interval; PPI, proton pump inhibitor; PR, prospective cohort; RC, retrospective cohort; RR, risk ratio.

Discussion
Results of this meta-analysis showed that PPI use was associated with moderately increased mortality in patients with cirrhosis. The association between PPI use and mortality in cirrhotic patients seemed to be independent of demographic factors and scales indicating the disease severity because we included data after adjustment of these confounding factors. Moreover, sensitivity analyses by omitting one study at a time indicated the robustness of the outcome. Subgroup analyses also suggested that the association between PPI use and mortality in cirrhotic patients was consistent in different subgroups according to study characteristics including patient ethnicity, sample size, with or without patient complications, definition of PPI use, and follow-up durations. Although the association between PPI use and mortality was significant in subgroup of retrospective studies but not in subgroups of prospective studies, the difference between the subgroups was not significant. Taken together, these results indicated that PPI use may be associated with increased mortality risk in patients with cirrhosis. Although these findings should be validated in large-scale prospective studies, considering the potential overutilization of PPI in cirrhotic patients, carefully selection of cirrhotic patients with appropriate indications for PPI therapy is recommended before using the medication. An early meta-analysis published in 2016 failed to show a significant association between PPI use and mortality risk in cirrhotic patients [11]. As mentioned by the authors, the limited number of available studies and sample size may lead to statistical inadequacy of the meta-analysis to reach a positive outcome. Our meta-analysis, by including 21 available cohort studies with 20,899 patients, showed that PPI use was associated with increased mortality in cirrhotic patients. Despite of the large number of studies included, our study has the following strengths. First, we only included article published in peer-reviewed journal, which therefore is of low risk of selection bias. Second, RRs data after adjustment of multiple confounding factors were included and combined, which therefore may provide an independent association between PPI use and mortality risk in cirrhosis patients. Third, we performed exhausting sensitivity and subgroup analyses to evaluate the potential influences of individual study and study characteristics on the outcome. These studies confirmed the robustness of the meta-analysis result, which was not likely to be driven by a single study or affected by study characteristics including patient ethnicity, sample size, with or without complications, definition of PPI, and follow-up durations. Although we found that the association between PPI use and mortality risk was significant in retrospective studies but not in prospective studies, difference between the subgroups was not statistically. Moreover, only five prospective cohorts were available. Therefore, it was likely that the non-significant association between PPI use and mortality risk in subgroup of prospective cohorts may probably be attributed to limited studies available. Since retrospective cohort studies is exposed to recall bias [41], large-scale prospective cohort studies are needed to validate our findings. Collectively, results of this meta-analysis support that PPI is associated with increased mortality in cirrhotic patients. Considering the potential overutilization of PPIs in patients with cirrhosis [5], as well as the potential safety consideration related to the long-term use of PPIs [3], decision for the prescription of PPIs in cirrhotic patients should only be made after carefully evaluating the appropriate indications and balancing the benefits and potential risk of adverse outcomes in these patients.
The reasons and mechanisms for the increased mortality risk associated with PPI use in cirrhotic patients may be multifactorial. First, PPIs, by suppressing gastric acids, may cause bacterial overgrowth in both stomach and intestine [42,43]. Subsequently, the impaired intestinal epithelial barrier in patients with cirrhosis [44] may facilitate the translocation of the overgrown bacteria to mesenteric lymph nodes, which finally leads to SBP and other infections [45]. Moreover, PPIs may also contribute to the gut dysbiosis that generally exists in patients with cirrhosis, whereas altered gut microbiota can induce or exacerbate HE [46,47], another complication that has been related to increased mortality risk in cirrhotic patients [12]. In addition, studies in overall population showed that PPI use may be associated with increased risks of cardiovascular adverse events [48], stroke [49], and fractures [50], which may also lead to increased mortality. Studies regarding the safety of PPI use in various populations, particularly the long-term use of PPI, should be performed to confirm these mechanisms.
Our study has limitations which should be considered when interpreting the results. First, significant heterogeneity exists among the included studies. Although exploring subgroup analyses were performed to evaluate the potential source of heterogeneity, the results of subgroup analyses should be interpreted with caution since the numbers of available studies were limited for each stratum of subgroups. Moreover, since the individual patient data were not available, we could only perform subgroup analyses based on study-level data. The influences of patient and study characteristics on the association between PPI use and mortality in cirrhotic patients should be further analyzed in future studies. Second, although we included studies with adjusted data for the association between PPI use and mortality, we could not exclude the existence of residual factors which may confound the association. Third, a causative association between PPI use and mortality in cirrhotic patients should not be derived based on our finding since the present study was a meta-analysis of observational studies. Randomized controlled trials should be considered to validate the influence of PPI on clinical outcomes in cirrhosis patients. Finally, we could not determine whether the dosages, durations, or individual PPI medications may affect the clinical outcomes of cirrhotic patients differently. Future studies are also warranted in this regard.