Background

Over four decades ago, investigators first described stress ulcer bleeding in critically ill patients [1]. Since then, multiple studies have described this condition and its impact on the prognosis of critically ill patients. Stress ulcers typically occur in the gastric body, esophagus, or duodenum, sometimes resulting in gastrointestinal (GI) bleeding. Earlier studies reported overt GI bleeding in 5 to 25 % of critically ill patients [2, 3]. In contrast, the incidence of clinically important GI bleeding is much lower, estimated between 1 and 4 % [2, 4–7]. A recent large observational study (1034 patients, 97 sites), reported a 2.6 % incidence of clinically important GI bleeding [7], which was previously found to be associated with increased intensive care unit (ICU) mortality and length of stay [8]. Despite reduction in clinically important GI bleeding, pharmacologic stress ulcer prophylaxis does not seem to affect mortality in randomized controlled trials (RCTs) [8].

RCTs have investigated different classes of medication for stress ulcer prophylaxis. Recently, a meta-analysis of 29 RCTs showed that prophylaxis with either proton pump inhibitors (PPIs) or histamine-2-receptor antagonists (H2RAs) was associated with lower risk of overt GI bleeding compared to placebo or no prophylaxis [9]. However, the relative effectiveness of the two classes of agent remains uncertain.

PPIs, more potent at increasing gastric pH than H2RAs and maintaining gastric pH between 3.5 and 5.0, may minimize the risk of gastric mucosal injury [10]. Of four meta-analyses comparing PPIs to H2RAs, three suggested that PPIs are superior to H2RAs [11–13] and one did not [14].

The Surviving Sepsis Campaign (SSC) guidelines recommend using stress ulcer prophylaxis among critically ill patients with risk factors (e.g., mechanically ventilated patients, and patients with coagulopathy), including a weak recommendation for using PPIs over H2RAs in this setting [15]. The advice is concordant with current practice: recent observational studies showed that PPIs are the most commonly used prophylactic agents in the ICU [16–18].

In terms of the relative impact of PPIs and H2RAs, adverse effects are also a concern. In particular, a recent large retrospective observational study suggested PPI versus H2RA use in critically ill patients was associated with higher risks of pneumonia and Clostridium difficile infection compared to H2RA [19]. These results are, however, limited by the observational study design.

Several RCTs have been published recently and may influence both risk of bias and precision [20–25]. Therefore, we conducted a systematic review and meta-analysis to evaluate the efficacy and safety of PPIs compared to H2RAs for stress ulcer prophylaxis in critically ill patients. We used the Grading of Recommendations Assessment, Development and Evaluation (GRADE) methodology to assess the quality of evidence [26].

Methods

Study selection

Studies were eligible if: (1) the study design was an RCT; (2) the population involved adult critically ill patients in the ICU; (3) the intervention group received a PPI (either parenteral or enteral), regardless of the dose, frequency, or duration; (4) the control group received an H2RA, either parenteral or enteral, regardless of the dose, frequency, or duration; and (5) the outcomes included all or any of the following: clinically important GI bleeding; overt upper GI bleeding; pneumonia; mortality, ICU length of stay, and/or Clostridium difficile infection.

Search strategy

We updated our previous systematic review [12] and searched MEDLINE, EMBASE, Cochrane Library, ACPJC, and International Clinical Trial Registry Platform (ICTRP) from March 2012 through November 2015. Our search strategy is detailed in Additional file 1: Tables S3-S5. We screened citations of all new potentially eligible articles without language or publication date restrictions. We conducted an electronic search of conference proceedings via a website provided by McMaster University (http://library.mcmaster.ca/articles/proceedingsfirst). Two reviewers (FA and EB) screened titles and abstracts to identify articles for full review, and evaluated the full text of potentially eligible studies. Disagreements between reviewers were resolved by consensus, and if necessary, consultation with a third reviewer (WA).

Data extraction

Two reviewers (FA and EB) independently extracted pertinent data from all new studies utilizing a pre-designed data abstraction form. Disagreements were resolved by discussion and consensus. We contacted study authors for missing or unclear information.

Risk of bias assessment

Two reviewers (FA and EB) independently examined eligible trials for risk of bias using the Cochrane Collaboration tool [27]. For each included trial, we judged articles as having low, unclear, or high risk of bias for the domains of adequate sequence generation, allocation sequence concealment, blinding for objective outcomes, incomplete outcome data, selective outcome reporting, and for other bias. The overall risk of bias for each trial included was categorized as low if the risk of bias was low in all domains, unclear if the risk of bias was unclear in at least one domain and with no high risk of bias domain, or high if the risk of bias was high in at least one domain. We resolved disagreements by discussion and consensus.

Statistical analysis

We analyzed data using RevMan software (Review Manager, version 5.3. Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration, 2014). We used the DerSimonian and Laird [28] random-effects model to pool the weighted effect of estimates across all studies. We estimated study weights using the inverse variance method. We calculated pooled relative risks (RRs) for dichotomous outcomes and mean differences (MDs) for continuous outcomes, with corresponding 95 % confidence intervals (CIs). We assessed statistical heterogeneity using Chi2 and I 2 statistics [29]. We predefined substantial heterogeneity as P < 0.10 or I 2 > 50 %.

We calculated the number needed to treat (NNT) using the method proposed by the Cochrane Collaboration [30]. We used an assumed control group (ACR) event rate of 3 % for clinically important bleeding and 5 % for overt GI bleeding; these ACRs were based on the results of a recent observational study [7]. We inspected funnel plots and performed Egger’s test to assess publication bias [31]. We explored heterogeneity between studies by performing predetermined subgroup analyses to investigate whether certain factors influenced the treatment effect. These subgroups included high vs. low risk of bias (hypothesizing that studies with high risk of bias would have a larger treatment effect), PPI route of administration (hypothesizing that the treatment effect would be larger with parenteral administration), PPI dose (hypothesizing that treatment effect would be larger with higher dosing). In addition, we conducted a post hoc sensitivity analysis, excluding trials published in abstract form [20, 23, 32–35].

Results

Characteristics of studies included

Our new search identified a total of 255 citations. After removing duplicates, 214 articles remained. Following screening of titles and abstracts, 197 articles were excluded; 17 articles were retrieved for full text assessment and 11 were excluded for variable reasons (Fig. 1). After reviewing our previous results, we excluded an abstract [36] (n = 202) that was later published as a full article [35]. Another study was published as an abstract but was excluded from the analysis because the necessary data could not be obtained [37]. A total of six new trials (n = 600 patients) are included in the different analyses.

Fig. 1
figure 1

Process of identifying eligible studies: 18 trials (5 abstracts and 13 full published articles) were eligible and were included in the qualitative and quantitative analyses. RCT randomized controlled trial

Full size image

Combining our previous and current results, 19 RCTs [20, 22–25, 32–35, 38–48] from 20 reports (one study published outcomes separately in two different reports) [47, 48] met eligibility criteria and were included. Two eligible trials were published in abstract form [32, 33]; further information was obtained after contacting the authors.

Of 19 eligible trials [20, 22–25, 32–35, 38–48], 6 were published as an abstract only [20, 23, 32–34, 38] (Table 1). Overall, the included RCTs enrolled 2117 critically ill patients with a wide spectrum of medical and surgical conditions. Ten trials used intravenous PPIs, and eight used enteral PPIs, and the route was not described in one trial, which was published in abstract form. [23] The definitions for bleeding and pneumonia varied across trials and are summarized in Table 1.

Table 1 Characteristics of trials included
Full size table

Risk of bias assessment

Using the Cochrane risk of bias tool, three trials were judged to be at low risk of bias, and for six trials the risk of bias was unclear (Additional file 1: Table S6). We could not evaluate the risk of bias in six trials published as abstracts [20, 23, 32–34, 38]. In total, 10 trials were judged to be at high risk of bias, primarily due to lack of or inappropriate blinding.

Assessment of quality of the evidence

We used the GRADE method [26] to assess the quality of evidence for individual outcomes. We present the details of our assessment in Table 2.

Table 2 Grading of Recommendation, Assessment, Development, and Evaluation (GRADE) evidence profile
Full size table

Main outcomes

A total of 14 trials enrolling 1679 patients reported clinically important GI bleeding (Fig. 2). PPI use was associated with lower risk of clinically important GI bleeding compared to H2RAs (RR 0.39; 95 % CI 0.21, 0.71; P = 0.002; I 2 = 0 %; moderate confidence). Using an assumed control event rate of 3 %, the number needed to treat (NNT) was 55 (95 % CI 42, 115). Seventeen trials enrolling 1897 patients reported overt GI bleeding (Fig. 3). Prophylaxis with PPI was associated with a lower risk of overt GI bleeding compared to H2RA (RR 0.48; 95 % CI 0.34, 0.66; P < 0.0001; I 2 = 3 %, moderate confidence). The NNT to prevent GI bleeding was 37 (95 % CI 29, 59) for an assumed control event rate of 5 %. Thirteen trials enrolling 1571 patients reported the risk of pneumonia (Fig. 4). The risk of pneumonia was similar between groups (RR 1.12; 95 % CI 0.86, 1.46; P = 0.39; I 2 = 2 %, low confidence). Eleven trials enrolling 1487 patients reported on mortality (Additional file 1: Figure S5). Mortality risk was similar between groups (RR 1.05; 95 % CI 0.87, 1.27; P = 0.61; I 2 = 0 %, moderate confidence). Seven trials enrolling 744 patients reported ICU length of stay (Additional file 1: Figure S6); ICU length of stay was not significantly different between groups (MD –0.38 days; 95 % CI –1.49, 0.74; P = 0.51; I 2 = 30 %, low confidence). None of the RCTs included reported on Clostridium difficile infection.

Fig. 2
figure 2

Clinically important gastrointestinal bleeding. Data from 14 trials (n = 1679 patients) are included, analyzed using the random effects model. Proton pump inhibitors (PPIs) were associated with a significantly lower risk of clinically important bleeding compared to histamine-2-receptor antagonists (H2RAs). IV Inverse Variance

Full size image
Fig. 3
figure 3

Overt upper gastrointestinal bleeding. Data from 17 trials (n = 1897 patients) are included, analyzed using the random effects model. Proton pump inhibitors (PPIs) were associated with a significantly lower risk of overt bleeding compared to histamine-2-receptor antagonists (H2RAs). IV Inverse Variance

Full size image
Fig. 4
figure 4

Pneumonia outcome. Data from 12 trials (n = 1471 patients) were included, analyzed using the random effects model. The risk of pneumonia was similar in both groups. PPI proton pump inhibitor, H2RA histamine-2-receptor antagonist. IV Inverse Variance

Full size image

Subgroup analyses

We found no statistically significant interaction between the magnitude of effect and risk of bias, route of PPI administration, or frequency of PPI dosing. Details of the results of the subgroup analyses are in Additional file 1: Table S4 and S5.

Sensitivity analysis excluding trials published as abstracts yielded results very similar to the primary analysis (RR 0.42; 95 % CI 0.21, 0.82; P = 0.01; I 2 = 0 %) (Additional file 1: Figure S7).

Publication bias

Visual inspection of the funnel plot for clinically important GI bleeding did not suggest the presence of publication bias (Additional file 1: Figure S8). The Egger test also supported this conclusion (–0.69; 95 % CI –2.44, 0.84; P = 0.28). The Egger test was significant for overt GI bleeding, which may suggest the presence of publication bias (–0.87; 95 % CI –1.67, –0.07; P = 0.03). We did not find evidence of publication bias for the outcomes of mortality and pneumonia (Additional file 1: Figure S9 and S10).

Discussion

This systematic review demonstrated moderate quality evidence that PPIs are superior to H2RAs in reducing the risk of both clinically important and overt GI bleeding. The relative treatment effect was large (relative risk reduction of 61 % for clinically important GI bleeding), and the NNT was 55, which translates into 16 fewer bleeding events per 1000 patients. The relatively low incidence of GI bleeding currently seen in the ICU explains the apparent discrepancy between a large relative effect and small absolute effect (Table 2).

The primary concern associated with PPI use in the ICU is the potentially higher risk of infection, particularly, pneumonia and C. difficile [19], potentially a consequence of attenuation of the gastric acid protection against bacteria. Patients with achlorhydria or on long-term acid suppressive therapy have increased bacterial growth on gastric mucosal biopsy [49]. Whether this translates into increased risk of infection in critically ill patients remains unknown.

Our meta-analysis did not suggest an increased risk of pneumonia associated with PPIs rather than H2RAs. Furthermore, mortality and duration of stay in the ICU did not differ between groups. The impact of acid suppressive therapy on Clostridium difficile infection is uncertain as none of the included trials reported on this. A systematic review of 12 observational studies evaluating 2948 non-ICU patients with Clostridium difficile found an association with acid suppressive therapy (OR 1.94; 95 % CI 1.37, 2.75). This association was larger with PPI use (OR 2.05; 95 % CI 1.47, 2.85) as compared to H2RA (OR 1.47; 95 % CI 1.06, 2.05), but the difference was not statistically significant (P = 0.17) [50]. Furthermore, a retrospective cohort study that used propensity matching and included over 30,000 critically ill patients suggested that PPI use was associated with a small increase in the risk of Clostridium difficile infection in comparison to H2RA (3.4 % vs. 2.7 %; P = 0.02) [19]. As these results are limited by risk of bias associated with observational designs and imprecision, randomized trials are needed to confirm or refute these observations.

Our meta-analysis used broad eligibility criteria to enhance the generalizability of the results. Despite including a wide spectrum of critically ill patients, there was no significant heterogeneity across trials for these outcomes. Our findings suggest that PPIs are effective in preventing stress ulcer bleeding without increasing the risk of pneumonia or mortality. Nonetheless, several factors suggest cautious interpretation of these results. The quality of evidence was moderate for most outcomes; this is primarily related to risk of bias. Nine trials did not employ proper blinding of healthcare providers or outcome assessors (Additional file 1: Table S6), which may have inflated the observed treatment effect of PPIs. Furthermore, a subgroup analysis based on risk of bias (low vs. high or unclear) suggested a larger treatment effect in trials of lower quality, although the test for interaction was not significant. The outcome definitions varied across studies, which may have affected the estimate of effect.

Other factors, such as enteral nutrition, may modify the efficacy of prophylactic PPIs and H2RAs. One small RCT in critically ill patients with burns showed that enteral nutrition increased gastric blood flow [51]. No RCTs exclusively examined the effect of enteral feeding on bleeding from stress ulcers. Whether enteral feeding modifies the effect of PPIs on the risk of bleeding or pneumonia remains uncertain.

Prior meta-analyses examined the effect of PPIs compared to H2RAs for stress ulcer prophylaxis [11–14]. Our meta-analysis included more trials than other meta-analyses on this topic (19 trials enrolling 2117 patients), improving the precision of our findings. We obtained additional missing data from the authors of the original trial reports to inform the analyses, and conducted subgroup analyses to assess the robustness of the findings. Using the GRADE approach to assess our confidence in the estimates of treatment effect, the certainty of evidence was moderate for clinically important and overt upper GI bleeding, and mortality outcomes, while it was low for pneumonia and ICU length of stay outcomes. We also adhered to the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) reporting guidelines [52].

Conclusions

In summary, our meta-analysis provides moderate quality evidence for clinicians and guideline groups suggesting that PPIs, when compared to H2RAs, lower the risk of clinically important and overt GI bleeding among critically ill patients, without increasing the risk of pneumonia and mortality, or ICU length of stay. The impact of these drugs on the risk of Clostridium difficile infection has yet to be examined in randomized trials in the ICU setting.