Participants, interventions, and comparisons

The 107 studies randomised a total of 15,057 participants to 27 comparisons involving 14 different treatment modalities. Most studies specifically mentioned that they randomised individuals who had no history of GI haemorrhage or peptic ulcer or gastritis or were not undergoing treatments for any of these conditions. The included studies included participants admitted to intensive care units, although the level of ICU into which participants were admitted was not clearly mentioned. Neither was a duration of ICU admission ≥ 48 hours ‐ a necessary inclusion criterion across studies. Among the included studies were five exclusively paediatric studies (Behrens 1994; Kuusela 1997; Lacroix 1986; Lopez‐Herce 1992; Yildizdas 2002); six quasi‐randomised studies (Borrero 1984; Borrero 1985; Borrero 1986; Brophy 2010; Martin 1980; Weigelt 1981); and seven studies reported as conference abstracts only (Fink 2003; Fogas 2013; Larson 1989; Luk 1982; Phillips 1998; Selvanderan 2016; Wee 2013).

Funding sources

Most study reports that included information about funding sources mentioned that researchers had received funding from pharmaceutical companies that were involved in production of the tested interventions (23 studies). A total of 17 studies reported institutional funding from, for example, the hospitals involved in the study. Sixty studies provided no information about funding.

Ongoing studies

We retrieved 12 ongoing studies that might be relevant to this review and will be included in the next update.

Studies awaiting classification

We were unable to obtain the full text of two studies and classified them as studies awaiting classification (Labattut 1992; Morris 2001).

We retrieved from study registries 12 studies that were ongoing or finished with no publication identified (ACTRN12616000481471; EUCTR2015‐000318‐24‐DK; EudraCT 2007‐006102‐19; IRCT201104134578N2; ISRCTN12845429; Krag 2016; NCT00590928; NCT00702871; NCT02157376; NCT02290327; NCT02718261; NCT03098537).

Performance bias

Thirty‐seven studies clearly mentioned the methods employed for blinding participants and personnel across all outcomes, and so review authors judged their risk of performance bias as low (Barandun 1985; Bashar 2013; Bonten 1995; Burgess 1995; Chan 1995; Conrad 2005; Cook 1998; Ephgrave 1998; Fogas 2013; Friedman 1982; Groll 1986; Halloran 1980; Hanisch 1998; Kappstein 1991; Khorvash 2014; Krakamp 1989; Kuusela 1997; Lacroix 1986; Larson 1989; Lee 2014; Luk 1982; Martin 1992; Martin 1993; Metz 1993; Ng 2012; Pan 2004; Peura 1985; Pickworth 1993; Powell 1993; Selvanderan 2015; Selvanderan 2016; Stothert 1980; Tabeefar 2012; van den Berg 1985; van Essen 1985; Wee 2013; Zinner 1989).

Detection bias

Eighty studies reported that researchers used a clear definition for detection of upper GI bleeding, and review authors judged these studies as having low risk of detection bias for this outcome (Ali 2016; Apte 1992; Basso 1981; Ben‐Menachem 1994; Borrero 1984; Borrero 1985; Borrero 1986; Bresalier 1987; Brophy 2010; Burgess 1995; Cannon 1987; Chan 1995; Cioffi 1994; Conrad 2005; Cook 1998; Darlong 2003; Davies 2012; Driks 1987; Eddleston 1991; Eddleston 1994; Ephgrave 1998; Fabian 1993; Fink 2003; Friedman 1982; Groll 1986; Halloran 1980; Hanisch 1998; Kantorova 2004; Kappstein 1991; Karlstadt 1990; Kaushal 2000; Kingsley 1985; Kitler 1990; Krakamp 1989; Kuusela 1997; Lacroix 1986; Laggner 1988; Lamothe 1991; Larson 1989; Lee 2014; Levy 1997; Lin 2016; Lopez‐Herce 1992; Luk 1982; Macdougall 1977; Maier 1994; Martin 1980; Martin 1992; Martin 1993; Metz 1993; Ng 2012; Ortiz 1998; Peura 1985; Pickworth 1993; Pinilla 1985; Poleski 1986; Prakash 2008; Priebe 1980; Prod'hom 1994; Reusser 1990; Rohde 1980; Ruiz‐Santana 1991; Ryan 1993; Selvanderan 2015; Selvanderan 2016; Skillman 1984; Solouki 2009; Somberg 2008; Stothert 1980; Terzi 2009; Thomason 1996; Tryba 1985; Tryba 1987; Tryba 1988; van den Berg 1985; van Essen 1985; Wang 2015; Weigelt 1981; Zinner 1981; Zinner 1989). Thirty‐eight studies reported that researchers used a clear definition for detection of nosocomial pneumonia, and review authors judged their risk of detection bias as low (Apte 1992; Bashar 2013; Ben‐Menachem 1994; Bonten 1995; Cioffi 1994; Conrad 2005; Cook 1998; Davies 2012; Driks 1987; Eddleston 1991; Eddleston 1994; Ephgrave 1998; Fabian 1993; Fogas 2013; Hanisch 1998; Israsena 1987; Kantorova 2004; Kappstein 1991; Khorvash 2014; Laggner 1989; Lee 2014; Lin 2016; Mahul 1992; Maier 1994; Martin 1993; Metz 1993; Mustafa 1994; Pickworth 1993; Prakash 2008; Prod'hom 1994; Ryan 1993; Sirvent 1994; Solouki 2009; Somberg 2008; Thomason 1996; Tryba 1987; Tryba 1988; Yildizdas 2002).

Clinically important upper GI bleeding

Clinically important upper GI bleeding was analysed (Analysis 1.1) for the comparison of any intervention (H2 receptor antagonists, proton pump inhibitors, prostaglandin analogues, anticholinergics, and sucralfate) versus placebo or no prophylaxis. We identified 31 studies that contributed data to this outcome, which we categorised according to drug class. There seems to be a beneficial effect of any intervention versus placebo or no prophylaxis on the occurrence of clinically important upper GI bleeding (risk ratio (RR) 0.47, 95% confidence interval (CI) 0.39 to 0.57; moderate certainty of evidence). We have elaborately mentioned further results below under each respective comparison.

Nosocomial pneumonia

Nosocomial pneumonia (which included ventilator‐associated pneumonia) was analysed (Analysis 1.2) for the comparison of any interventions (H2 receptor antagonists, proton pump inhibitors, anticholinergics, and sucralfate vs placebo or no prophylaxis). We identified nine studies that contributed data to this outcome and categorised these data according to drug class. With respect to the occurrence of differences in nosocomial pneumonia between study groups (RR 1.15, 95% CI 0.90 to 1.48; low certainty of evidence), the use of any intervention versus no intervention or prophylaxis was consistent with benefits and harms. We have elaborately mentioned these results below under each respective comparison.

All‐cause mortality in ICU

All‐cause mortality in the ICU was analysed (Analysis 1.3) for the comparison of any interventions (H2 receptor antagonists, proton pump inhibitors, prostaglandin analogues, anticholinergics, antacids, and sucralfate vs placebo or no prophylaxis). We identified 18 studies that contributed data to this outcome and organised the data from these studies according to drug class. With respect to all‐cause mortality in the ICU, the effect estimate of any intervention versus no intervention or prophylaxis was consistent with benefit and harm (RR 1.10, 95% CI 0.90 to 1.34; low certainty of evidence). We have elaborately mentioned these results below under each respective comparison.

All‐cause mortality in hospital

All‐cause mortality in the hospital was analysed (Analysis 1.4) for the comparison of any interventions (H2 receptor antagonists, proton pump inhibitors, antacids, and sucralfate vs placebo or no prophylaxis). We identified five studies that contributed data to this outcome and categorised the data from these studies according to drug class. The effect estimate of any intervention versus no intervention or prophylaxis was consistent with benefits and harms with respect to all‐cause mortality in the hospital (RR 1.15, 95% CI 0.85 to 1.55). We have elaborately mentioned these results below under each respective comparison.

Duration of ICU stay

Duration of ICU stay was analysed (Analysis 1.5) for the comparison of any interventions (H2 receptor antagonists, proton pump inhibitors, and sucralfate vs placebo or no prophylaxis). We identified two studies that contributed data to this outcome and categorised the data from these studies according to drug class. The effect estimate of any intervention versus no intervention or prophylaxis was consistent with benefits and harms with respect to duration of ICU stay (mean difference (MD) 0.24 days, 95% CI ‐1.13 to 1.61; low certainty of evidence). We have elaborately mentioned these results below under each respective comparison.

Duration of intubation

Duration of intubation was analysed (Analysis 1.6) for the comparison of any interventions (H2 receptor antagonists, proton pump inhibitors, and sucralfate vs placebo or no prophylaxis). We identified two studies that contributed data to this outcome and categorised the data from these studies according to drug class. The effect estimate of any intervention versus no intervention or prophylaxis was consistent with benefits and harms with respect to duration of intubation (MD 0.87 days, 95% CI ‐0.58 to 2.31; low certainty of evidence). We have elaborately mentioned these results below under each respective comparison.

Number of participants requiring blood transfusions

The number of participants requiring blood transfusions was analysed (Analysis 1.7) for the comparison of any interventions (H2 receptor antagonists, antacids, and sucralfate vs placebo or no prophylaxis). We identified nine studies that contributed data to this outcome and categorised the data from these studies according to drug class. Data seem to show a reduction in the number of participants in the intervention group requiring blood transfusion (RR 0.63, 95% CI 0.41 to 0.97; moderate certainty of evidence). We have elaborately mentioned these results below under each respective comparison.

Units of blood transfused

The number of units of blood transfused was analysed (Analysis 1.8) for the comparison of any interventions (H2 receptor antagonists and sucralfate vs placebo or no prophylaxis). We identified two studies that contributed data to this outcome and categorised the data from these studies according to drug class. The effect estimate of any intervention versus no intervention or prophylaxis was consistent with benefits and harms with respect to the number of units of blood transfused (MD 0.09 units, 95% CI ‐0.99 to 1.17). We have elaborately mentioned these results below under each respective comparison.

Clinically important upper GI bleeding

We identified 24 studies relevant to this outcome (N = 2149) and stratified data into six groups according to drug class members of H2 receptor antagonists. For this outcome, we found evidence suggesting that H2 receptor antagonists reduced the risk of clinically important upper GI bleeding compared with placebo or no prophylaxis (8.3% vs 17.7%; RR 0.50, 95% CI 0.36 to 0.70; Analysis 2.1). We judged the certainty of the evidence as moderate. This outcome had moderate levels of heterogeneity (Chi² = 38.78, df = 23.0 (P = 0.02), I² = 40%).

Nosocomial pneumonia

We identified eight studies that were relevant to this outcome (N = 945) and divided the data into five subcategories in accordance with our protocol. The effect of H2 receptor antagonists versus no prophylaxis on nosocomial pneumonia was consistent with benefits and harms (15.8% vs 14.6%; RR 1.12, 95% CI 0.85 to 1.48; Analysis 2.2). We judged the certainty of this evidence as low.

All‐cause mortality in ICU

We identified 14 studies that contributed data to this outcome (N = 1428). With respect to all‐cause mortality in the ICU, the effect of H2 receptor antagonists versus no prophylaxis on nosocomial pneumonia was consistent with benefits and harms (15.8% vs 14.3%; RR 1.12, 95% CI 0.88 to 1.42; Analysis 2.3). We judged the certainty of this evidence as low.

All‐cause mortality in hospital

For this outcome, we found four relevant studies (N = 487). Data seemed to show no difference between H2 receptor antagonists and placebo or no prophylaxis with respect to all‐cause mortality in the hospital (18.4% vs 15.7%; RR 1.16, 95% CI 0.79 to 1.70; Analysis 2.4).

Duration of ICU stay

We identified two studies relevant to this outcome (N = 230). Data seemed to show no difference between H2 receptor antagonists and placebo or no prophylaxis with respect to duration of ICU stay (MD 0.73 days, 95% CI ‐0.92 to 2.38; Analysis 2.5). We judged the certainty of this evidence as low.

Duration of intubation

For this outcome, we found two relevant studies and categorised data into two comparisons (N = 230). Data seemed to show no difference between H2 receptor antagonists and placebo or no prophylaxis with respect to duration of intubation (MD 0.79 days, 95% CI ‐0.95 to 2.54; Analysis 2.6). Apte 1992 assessed the duration of intubation but did not report enough data for inclusion in meta‐analyses.

Number of participants requiring blood transfusions

Data seemed to show a beneficial effect of H2 receptor antagonists compared with placebo or no prophylaxis with respect to the number of participants requiring blood transfusions (6.4% vs 11.2%; RR 0.58, 95% CI 0.36 to 0.95; seven studies; 655 participants; Analysis 2.7). We judged the certainty of this evidence as moderate.

Units of blood transfused

For this outcome, we found two relevant studies and categorised data into two comparisons (N = 209). Data seemed to show no difference between H2 receptor antagonists and placebo or no prophylaxis with respect to the number of units of blood transfused (MD 0.33 units, 95% CI ‐0.04 to 0.70; Analysis 2.8). This outcome had considerable heterogeneity (Chi² = 9.49, df = 1.0 (P = 0.0), I² = 89%).

Adverse events of interventions

For this outcome we found eight relevant studies and divided the data into 12 comparisons according to type of adverse event.

Clinically important upper GI bleeding

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 3.1). In the three included studies that contributed data to the analyses, 114 participants received proton pump inhibitors and 123 participants received either placebo or no prophylaxis. The event occurred in 2.6% and 4.9% of participants in the two groups. Data seemed to show no difference between proton pump inhibitors and placebo or no prophylaxis with respect to the occurrence of clinically important upper GI bleeding (RR 0.63, 95% CI 0.18 to 2.22). We judged the certainty of the evidence as low.

We identified two additional studies. Selvanderan 2016 included 214 participants and did not report how many participants were randomised to the two interventions. Powell 1993 included 21 participants in the proton pump inhibitor arm and 10 in the placebo arm. Both trials reported no events of clinically important upper GI bleeding in either group.

Nosocomial pneumonia

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 3.2). In the two included studies, 110 participants received proton pump inhibitors and 117 received placebo. Nosocomial pneumonia occurred in 22.8% of the group that received proton pump inhibitors and 18.7% of the group that received placebo or no prophylaxis. Data seemed to show no difference between proton pump inhibitors and placebo or no prophylaxis with respect to the occurrence of nosocomial pneumonia (RR 1.24, 95% CI 0.77 to 1.98). We judged the certainty of the evidence as low.

We identified an additional study that was published as a conference abstract and did not report enough data for formal meta‐analysis (Selvanderan 2016). Researchers randomised a total of 214 participants to receive pantoprazole or placebo. One participant in the placebo group and two in the pantoprazole group developed ventilator‐associated complications including pneumonia. Data showed eight and 12 cases of clinician‐adjudicated nosocomial pneumonia in pantoprazole and control groups, respectively.

All‐cause mortality in ICU

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 3.3). In the three included studies, 131 participants received proton pump inhibitors and 127 received either placebo or no prophylaxis. The event occurred in 13.3% and 13.4% of participants in the two groups. Data seemed to show no difference between proton pump inhibitors and placebo or no prophylaxis with respect to all‐cause mortality in the ICU (RR 1.09, 95% CI 0.60 to 1.99). We judged the certainty of this evidence as low.

We identified an additional study that was published as a conference abstract and did not provide enough data for formal meta‐analysis (Selvanderan 2016). Researchers randomised a total of 214 participants to receive pantoprazole or placebo. The study reported an adjusted hazard ratio for mortality with pantoprazole of 1.68 (95% CI 0.97 to 2.90).

All‐cause mortality in hospital

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 3.4). In the included study, 75 participants received proton pump inhibitors and 75 received placebo. All‐cause mortality occurred in 18.7% of the group that received proton pump inhibitors and 17.3% of the group that received placebo. Data seemed to show no difference between proton pump inhibitors and placebo or no prophylaxis with respect to all‐cause mortality in hospital (RR 1.08, 95% CI 0.54 to 2.13).

Duration of ICU stay

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 3.5). In the two included studies, 110 participants received proton pump inhibitors and 117 received either placebo or no prophylaxis. Data seemed to show no difference between proton pump inhibitors and placebo or no prophylaxis with respect to length of ICU stay (MD ‐0.03 days, 95% CI ‐1.63 to 1.58; two studies; 227 participants; Analysis 3.5). We judged the certainty of this evidence as low.

Duration of intubation

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 3.6). In the two included studies, 110 participants received sucralfate and 107 received either placebo or no prophylaxis. Data seemed to show no difference between proton pump inhibitors and placebo or no prophylaxis with respect to duration of intubation (MD 0.36 days, 95% CI ‐1.43 to 2.15).

Adverse events

One study including 214 participants compared pantoprazole versus placebo (Selvanderan 2015). Researchers tested 30 participants assigned to pantoprazole and 40 assigned to placebo for Clostridium difficile infection. One participant in the pantoprazole group was infected with C difficile.

Clinically important upper GI bleeding

Data seemed to show no difference between proton pump inhibitors plus sucralfate and placebo or no prophylaxis with respect to the occurrence of clinically important upper GI bleeding (20.0% vs 37.5%; RR 0.53, 95% CI 0.26 to 1.12; one study; 80 participants who contributed data to this comparison (Analysis 4.1). We judged the certainty of this evidence as moderate.

Clinically important upper GI bleeding

Data seemed to show no difference between prostaglandin analogues and placebo with regard to the occurrence of clinically important upper GI bleeding (10.3% in both groups; RR 1.00, 95% CI 0.22 to 4.55; one study; 58 participants; Analysis 5.1). We judged the certainty of this evidence as moderate.

All‐cause mortality in ICU

Data seemed to show no difference between prostaglandin analogues and placebo with regard to all‐cause mortality in the ICU (27.6% and 24.1%; RR 1.14, 95% CI 0.48 to 2.74; one study; 58 participants; Analysis 5.2). We judged the certainty of this evidence as moderate.

Clinically important upper GI bleeding

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 6.1). In the two studies that contributed data towards this outcome, the event occurred in 10.2% and 9.7% of participants. Data seemed to show no difference between anticholinergics and placebo or no prophylaxis with respect to the occurrence of clinically important upper GI bleeding (RR 0.95, 95% CI 0.36 to 2.51). We judged the certainty of this evidence as low.

Nosocomial pneumonia

We classified the study on the basis of its intervention and comparison arms (as shown in Analysis 6.2). In the single included study, 44 participants received anticholinergics and 57 received placebo. The event occurred in 22.7% and 21.1% of participants in both groups. Data seemed to show no difference between anticholinergics and placebo with respect to the occurrence of nosocomial pneumonia (RR 1.08, 95% CI 0.51 to 2.27; one study; 101 participants). We judged the certainty of this evidence as low.

All‐cause mortality in ICU

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 6.3). In the two included studies, 59 participants received anticholinergics and 72 received placebo alone or placebo with ranitidine. The event occurred in 25.4% and 20.8% of participants in both groups. Data seemed to show no difference between anticholinergics and placebo with respect to all‐cause mortality in the ICU (RR 1.23, 95% CI 0.66 to 2.30). We judged the certainty of this evidence as low.

Duration of ICU stay

We identified one study that looked at this outcome but did not report enough data for inclusion in formal meta‐analyses (Hanisch 1998). In the included study, 44 participants received pirenzepine and 57 received placebo. The mean duration of ICU stay was 9.9 days (2 to 39) and 12.6 days (2 to 58), respectively.

Duration of intubation

We identified one study that looked at this outcome but did not report enough data for inclusion in formal meta‐analyses (Hanisch 1998). In the included study, 44 participants received pirenzepine and 57 received placebo. The mean duration of intubation was 8.0 days (2 to 32) and 10.2 days (2 to 55), respectively.

Units of blood transfused

We identified one study that looked at this outcome but did not report enough data for inclusion in formal meta‐analyses (Hanisch 1998). In the included study, 44 participants received pirenzepine and 57 received placebo. The mean number of units of blood transfused was four units in each of the three patients with GI bleeding in the pirenzepine group and 12 units in the single surviving participant with GI bleeding in the placebo group.

Clinically important upper GI bleeding

Data seemed to show a beneficial effect of antacids compared with placebo or no prophylaxis with respect to the occurrence of clinically important upper GI bleeding. The outcome occurred in 7.2% of the antacids group and in 17.7% of the placebo group (RR 0.49, 95% CI 0.25 to 0.99; eight studies; 774 participants; Analysis 7.1). We judged the certainty of this evidence as low. This outcome had substantial heterogeneity (Chi² = 15.92, df = 7.0 (P = 0.03), I² = 56%).

All‐cause mortality in ICU

Data seemed to show no difference between anticholinergics and no prophylaxis with respect to all‐cause mortality in the ICU. The event occurred in 15.9% of the antacids group and in 16.1% of the group that received no prophylaxis (RR 1.01, 95% CI 0.53 to 1.96; two studies; 300 participants; Analysis 7.2). We judged the certainty of the evidence as low.

All‐cause mortality in the hospital

Data seemed to show no difference between anticholinergics and no prophylaxis with respect to all‐cause mortality in the hospital. The event occurred in 30.7% of the intervention group and in 21.3% of the control group (RR 1.44, 95% CI 0.79 to 2.64; one study; 126 participants; Analysis 7.3).

Duration of intubation

We identified one study that did not report enough data for inclusion in formal meta‐analysis (Pinilla 1985). Study participants received either antacids (n = 65) or no specific prophylaxis (n = 61). The mean duration of ventilation in the antacids group was 136.4 hours (range 0 to 360) and in the control group 173.6 hours (range 12 to 552).

Number of participants requiring blood transfusions

Data seemed to show no difference between antacids and no prophylaxis with respect to the number of participants requiring blood transfusion. In the intervention group, 4.3% of participants required blood transfusion, and in the control group, 4.5% required blood transfusion (RR 0.94, 95% CI 0.30 to 2.96; two studies; 226 participants; Analysis 7.4). We judged the certainty of the evidence as low.

Adverse events of interventions

We identified four studies relevant to this outcome and categorised data according to seven adverse events.

Clinically important upper GI bleeding

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 8.1). Data seemed to show a beneficial effect of sucralfate compared with placebo or no prophylaxis with respect to the occurrence of clinically important upper GI bleeding. The event occurred in 6.3% of participants in the intervention group and in 10.8% of participants of the control group (RR 0.53, 95% CI 0.32 to 0.88; seven studies; 598 participants; Analysis 8.1). We judged the certainty of this evidence as moderate.

Nosocomial pneumonia

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 8.2). We identified four studies relevant to this outcome (N = 450). Data seemed to show no difference between sucralfate and placebo or no prophylaxis with respect to the occurrence of ventilator‐associated pneumonia. The event occurred in 15.9% and 12.2% of the two groups, respectively (RR 1.33, 95% CI 0.86 to 2.04; Analysis 8.2). We judged the certainty of this evidence as low.

All‐cause mortality in ICU

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 8.3). Data seemed to show no difference between sucralfate and placebo or no prophylaxis with respect to all‐cause mortality in the ICU. The event occurred in 16.3% and 16.5% in the two groups, respectively (RR 0.97, 95% CI 0.66 to 1.43; five studies; 500 participants). We judged the certainty of this evidence as low.

All‐cause mortality in hospital

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 8.4). Data seemed to who no difference between sucralfate and placebo or no prophylaxis with respect to all‐cause mortality in the hospital. The event occurred in 17.8% and 18.3% in the two groups, respectively (RR 0.97, 95% CI 0.62 to 1.52; two studies; 344 participants).

Duration of ICU stay

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 8.5). Data seemed to show no difference between sucralfate and placebo or no prophylaxis with respect to duration of ICU stay (MD ‐0.02 days, 95% CI ‐1.70 to 1.65; two studies; 224 participants). We judged the certainty of this evidence as low.

Duration of intubation

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 8.6). Data seemed to show no difference between sucralfate and placebo or no prophylaxis with respect to duration of intubation (MD 1.42, 95% CI ‐0.27 to 3.10; two studies; 224 participants).

Number of participants requiring blood transfusion

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 8.7). Data seemed to show no difference between sucralfate and no prophylaxis with respect to the number of participants requiring blood transfusion. Blood transfusion was required by 3% and 5% of the two groups, respectively (RR 0.60, 95% CI 0.15 to 2.44; one study; 200 participants). We judged the certainty of this evidence as low.

Units of blood transfused

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 8.8). Data seemed to show an unfavourable effect of sucralfate with respect to the number of units of blood transfused (MD 0.80, 95% CI 0.32 to 1.28; one study; 200 participants).

Adverse events of interventions

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 8.9). For this outcome, we found a single study that measured nausea and vomiting as an adverse event. Data seemed to show no difference between sucralfate and placebo or no prophylaxis with respect to the occurrence of adverse events. The sucralfate group included four participants with nausea and vomiting and the placebo group included none (RR 9.00, 95% CI 0.50 to 161.13; one study; 70 participants).

Clinically important upper GI bleeding

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 9.1). Data seemed to show increased risk of clinically important upper GI bleeding under H2 receptor antagonists when compared with proton pump inhibitors. The event occurred in 6.9% and 1.7% of participants in the two groups, respectively (RR 2.90, 95% CI 1.83 to 4.58; 13 studies; 1636 participants). Five studies with 525 participants reported that no events occurred in either group. We judged the certainty of this evidence as low.

Nosocomial pneumonia

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 9.2). Data seemed to show no difference between H2 receptor antagonists and proton pump inhibitors with respect to the occurrence of nosocomial pneumonia. The event occurred in 13.3% and 12.1% of participants in the two groups, respectively (RR 1.02, 95% CI 0.77 to 1.35; eight studies; 1256 participants). We judged the certainty of this evidence as low.

All‐cause mortality in ICU

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 9.3). Data seemed to show no difference between H2 receptor antagonists and proton pump inhibitors with respect to all‐cause mortality in the ICU. The event occurred in 19.3% and 16.3% of participants in the two groups, respectively (RR 0.96, 95% CI 0.78 to 1.19; 12 studies; 1564 participants). We judged the certainty of this evidence as low.

All‐cause mortality in the hospital

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 9.4). Data seemed to show no difference between H2 receptor antagonists and proton pump inhibitors with respect to all‐cause mortality in the hospital. The event occurred in 5.5% and 7.2% of participants, respectively (RR 0.72, 95% CI 0.37 to 1.43; two studies; 454 participants).

Duration of ICU stay

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 9.5). Data seemed to show no difference between H2 receptor antagonists and proton pump inhibitors with respect to duration of ICU stay (MD 0.14 days, 95% CI ‐1.14 to 1.41; five studies; 482 participants). We judged the certainty of this evidence as low.

Duration of intubation

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 9.6). We identified six studies relevant to this outcome (N = 671). However, only five studies contributed data to our meta‐analysis. We found no difference between H2 receptor antagonists and proton pump inhibitors with respect to duration of intubation (MD ‐0.35 days, 95% CI ‐1.48 to 0.78; five studies; 542 participants).

Number of participants requiring blood transfusion

Data seemed to show no difference between H2 receptor antagonists and proton pump inhibitors with respect to the number of participants requiring blood transfusion (3.8% and 1.7%; RR 1.98, 95% CI 0.75 to 5.21; three studies; 575 participants; Analysis 9.7). We judged the certainty of this evidence as low.

Adverse events of interventions

For this outcome, we found five relevant studies and categorised data according to 12 adverse events.

Clinically important upper GI bleeding

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 10.1). In the 16 included studies, 884 participants received ranitidine, cimetidine, or famotidine, and 816 received antacids. Clinically important upper GI bleeding occurred in 8.4% and 8.6% of participants in both groups. Data seemed to show no difference between the two groups with respect to the occurrence of clinically important upper GI bleeding (RR 0.96, 95% CI 0.67 to 1.36; 16 studies; 1700 participants). We judged the certainty of this evidence as low.

Nosocomial pneumonia

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 10.2). In the four included studies, 288 participants received ranitidine or cimetidine, and 293 received antacids. Nosocomial pneumonia occurred in 29.2% and 28.0% of participants in the two groups, respectively. Data seemed to show no difference between the two groups with respect to the occurrence of the outcome of interest (RR 1.05, 95% CI 0.81 to 1.36). We judged the certainty of this evidence as low.

All‐cause mortality in ICU

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 10.3). In the 11 included studies, 472 participants received cimetidine, 33 received cimetidine plus pirenzepine, and 178 received ranitidine. Overall, 638 participants received antacids. Tryba 1985 administered pirenzepine as a concomitant medication to both cimetidine and antacid arms. All‐cause mortality occurred in 16.1% and 16.3% of the participants in both groups. Data seemed to show no difference between the two groups with respect to all‐cause mortality (RR 1.01, 95% CI 0.66 to 1.55). We judged the certainty of this evidence as very low.

All‐cause mortality in hospital

We classified the study on the basis of its intervention and comparison arms (as shown in Analysis 10.4). In the single included study, 80 participants received ranitidine and 81 received antacids. The event occurred in 33.8% and 39.5% of participants in the two groups. Data seemed to show no difference between the two groups with respect to all‐cause mortality in the hospital (RR 0.85, 95% CI 0.57 to 1.29).

Duration of intubation

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 10.5). In the three included studies, 64 participants received cimetidine and 57 received antacids. Data seemed to show no difference between the two groups with respect to duration of intubation (MD ‐0.81 days, 95% CI ‐3.85 to 2.23).

Number of participants requiring blood transfusions

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 10.6). In the six included studies, 379 participants received cimetidine or ranitidine and 365 received antacids. The event occurred in 8.2% and 3.0% of participants in both groups. Data seemed to show increased risk of blood transfusion in the H2 receptor antagonists arm compared with the antacids arm (RR 2.49, 95% CI 1.35 to 4.62). We judged the certainty of this evidence as moderate.

Number of units of blood transfused

We identified one study that addressed this outcome but did not report enough data for inclusion in a formal meta‐analysis (Priebe 1980). The study included 38 participants who received cimetidine and 37 who received antacids. Blood transfusions were required only for participants with GI bleeding, all of whom were included in the cimetidine group. It was reported that one participant received two units of blood and three participants received a total of four units of blood.

Adverse events of interventions

We identified 12 studies relevant to this outcome and divided the data into 12 groups according to adverse events.

Clinically important upper GI bleeding

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 11.1). In the 24 included studies, 1761 participants received ranitidine, cimetidine, or famotidine, and 1638 received sucralfate. Clinically important upper GI bleeding occurred in 7.6% and 6.6% of participants in the two groups, respectively. Data seemed to show no difference between the two groups with respect to the occurrence of clinically important upper GI bleeding (RR 1.10, 95% CI 0.87 to 1.41; 24 studies; 3316 participants). An additional study with 83 participants reported no events of clinically important upper GI bleeding in either treatment group (Pickworth 1993). We judged the certainty of this evidence as low.

Nosocomial pneumonia

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 11.2). In the 17 included studies, 1547 participants received ranitidine or cimetidine or famotidine, and 1460 received sucralfate. Nosocomial pneumonia occurred in 23.5% and 18.9% of participants in both groups. Data seemed to show increased risk of nosocomial pneumonia in the H2 receptor antagonist group compared with the sucralfate group (RR 1.22, 95% CI 1.07 to 1.40; 17 studies; 3041 participants). We judged the certainty of the evidence as moderate.

All‐cause mortality in ICU

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 11.3). In the 21 included studies (N = 3178), 1652 participants received cimetidine, ranitidine, or famotidine, and 1526 received sucralfate. All‐cause mortality occurred in 22.0% and 20.4% of participants in both groups. We found no difference between the two groups with respect to all‐cause mortality in the ICU (RR 1.09, 95% CI 0.95 to 1.24). We judged the certainty of this evidence as low.

All‐cause mortality in hospital

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 11.4). In the four included studies (N = 717), 366 participants received cimetidine, ranitidine, or famotidine, and 351 received sucralfate. All‐cause mortality in the hospital occurred in 22.7% and 20.2% of participants in both groups. Data showed no significant differences between the two groups (RR 1.14, 95% CI 0.86 to 1.50).

Duration of intubation

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 11.5). In the 10 included studies (N = 1751), 864 participants received cimetidine, ranitidine, or famotidine, and 887 received sucralfate. Data seemed to show no difference between the two groups with respect to duration of intubation (MD 0.22 days, 95% CI ‐1.55 to 2.00). This outcome had considerable heterogeneity (Chi² = 50.79, df = 9.0 (P = 0.0), I² = 82%).

Duration of ICU stay

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 11.6). In the six included studies (N = 1791), 904 participants received ranitidine or famotidine and 887 received sucralfate. Data seemed to show no difference between the two groups with respect to duration of ICU stay (MD 0.01 days, 95% CI ‐1.92 to 1.95). For this outcome, heterogeneity was considerable (Chi² = 28.48, df = 5.0 (P = 0.0), I² = 82%) and heterogeneity could not be explained by differences among the different subcomparisons. We judged the certainty of this evidence as very low.

Number of participants requiring blood transfusion

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 11.7). In the nine included studies (N = 1095), 603 participants received cimetidine or ranitidine and 492 received sucralfate. Blood transfusions were required by 4.5% and 3.5% of participants in the two groups. Data seemed to show no difference between study arms with respect to the number of participants requiring blood transfusion (RR 1.25, 95% CI 0.70 to 2.23). We judged the certainty of this evidence as low.

Units of blood transfused

We classified the single study that contributed data for this outcome from its intervention and comparison arms (as shown in Analysis 11.8) (Ben‐Menachem 1994). In the included study (N = 200), 100 participants received cimetidine and 100 received sucralfate. The mean number of units of blood transfused was 1.6 (1.3) and 2.0 (2.0). Data seemed to show no difference between the two groups with respect to the mean number of units of blood transfused (MD ‐0.40, 95% CI ‐0.87 to 0.07). An additional study did not provide enough data (no SE or SD reported) for inclusion in formal meta‐analysis (Fabian 1993). This study reported that the mean number of required blood transfusions was 9.0 in the sucralfate group (n = 206) and 10.3 in the cimetidine group (n = 410).

Adverse events of interventions

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 11.9). We identified six studies relevant to this outcome and divided the data from these studies according to eight adverse events.

Clinically important upper GI bleeding

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 12.1). In the three included studies (N = 556), 285 participants received either ranitidine or cimetidine, and 271 received pirenzepine. Clinically important upper GI bleeding occurred in 4.2% and 3.0% of participants in the two groups. Data seemed to show no difference between the two groups with respect to the occurrence of clinically important upper GI bleeding (RR 1.37, 95% CI 0.58 to 3.26). We rated the certainty of this evidence as low.

Nosocomial pneumonia

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 12.2). In the three included studies (N = 544), 279 participants received ranitidine or famotidine, and 265 received pirenzepine. Nosocomial pneumonia occurred in 6.5% and 5.3% of participants in both groups. Data seemed to show no difference between the two groups with respect to occurrence of the outcome of interest (RR 0.96, 95% CI 0.50 to 1.84). We judged the certainty of this evidence as low.

All‐cause mortality in ICU

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 12.3). In the two included studies (N = 501), 257 participants received ranitidine and 244 received pirenzepine. All‐cause mortality occurred in 5.8% and 6.6% of participants in the two groups. Data seemed to show no difference between the two groups with respect to occurrence of the outcome of interest (RR 0.89, 95% CI 0.21 to 3.87). This outcome had substantial heterogeneity (Chi² = 4.08, df = 1.0 (P = 0.04), I² = 75%). We judged the certainty of this evidence as very low.

Duration of ICU stay

We identified one study that looked at this outcome but did not report sufficient data for inclusion in formal meta‐analyses (Hanisch 1998). In the included study, 57 participants received ranitidine and 44 received pirenzepine. The mean duration of ICU stay was 9.7 days (2 to 95) and 9.9 days (2 to 39), respectively.

Duration of intubation

We identified one study that looked at this outcome but did not report sufficient data for inclusion in formal meta‐analyses (Hanisch 1998). In the included study, 57 participants received ranitidine and 44 received pirenzepine. The mean duration of mechanical ventilation was 8.2 days (2 to 93) and 8.0 days (2 to 32), respectively.

Number of participants requiring blood transfusions

We classified the study on the basis of its intervention and comparison arms (as shown in Analysis 12.4). In the included study, 57 participants received ranitidine and 44 received pirenzepine. The event occurred in 5.3% and 6.8% of participants in the two groups. Data seemed to show no difference between the two groups with respect to the number of participants requiring blood transfusion (RR 0.77, 95% CI 0.16 to 3.64). We judged the certainty of this evidence as low.

Number of units of blood transfused

We identified one study that looked at this outcome but did not report sufficient data for inclusion in formal meta‐analyses (Hanisch 1998). In the included study, 57 participants received ranitidine and 44 received pirenzepine. The mean number of units of blood transfused was two units for each of the three participants with GI bleeding in the ranitidine group, and four units for each of the four participants with GI bleeding in the pirenzepine group.

Adverse events of interventions

We classified studies on the basis of their intervention and comparison arms and adverse events of interventions (as shown in Analysis 12.5). Researchers reported no adverse events other than tachycardia and high temperature.

Clinically important upper GI bleeding

We classified this study on the basis of intervention and comparison arms (as shown in Analysis 13.1). In the single included study, 64 participants received cimetidine and 63 received misoprostol. Clinically important upper GI bleeding occurred in 4.7% and 11.1% of participants in the two groups. Data seemed to show no difference between the two groups with respect to the occurrence of clinically important upper GI bleeding (RR 0.42, 95% CI 0.11 to 1.56). We judged the certainty of this evidence as low.

All‐cause mortality in ICU

We classified this study on the basis of intervention and comparison arms (as shown in Analysis 13.2). In the included study, 64 participants received cimetidine and 63 received misoprostol. All‐cause mortality occurred in 39.1% and 30.2% of participants in the two groups. Data seemed to show no difference between the two groups with respect to all‐cause mortality (RR 1.30, 95% CI 0.80 to 2.10). We judged the certainty of this evidence as moderate.

Clinically important upper GI bleeding

We classified this study on the basis of intervention and comparison arms (as shown in Analysis 14.1). In the included study, 70 participants received ranitidine and 70 received teprenone. Clinically important upper GI bleeding occurred in 5.7% and 5.7% of participants in the two groups. Data seemed to show no difference between the two groups with respect to the occurrence of clinically important upper GI bleeding (RR 1.00, 95% CI 0.26 to 3.84). We judged the certainty of this evidence as low.

All‐cause mortality in ICU

We classified this study on the basis of intervention and comparison arms (as shown in Analysis 14.2). In the included study, 70 participants received ranitidine and 70 received teprenone. All‐cause mortality in the ICU occurred in 1.4% and 1.4% of participants in the two groups. Data seemed to show no difference between the two groups with respect to all‐cause mortality (RR 1.00, 95% CI 0.06 to 15.67). We judged the certainty of this evidence as moderate.

Number of participants requiring blood transfusions

We classified this study on the basis of intervention and comparison arms (as shown in Analysis 14.3). In the included study, 70 participants received ranitidine and 70 received teprenone. Blood transfusions were required by 5.7% and 5.7% of participants in both groups. Data seemed to show no difference between the two groups with respect to the number that needed blood transfusion (RR 1.00, 95% CI 0.26 to 3.84). We judged the certainty of this evidence as moderate.

Clinically important upper GI bleeding

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 15.1). In the two included studies that contributed data to this outcome, 119 participants received ranitidine or cimetidine plus antacids, and 111 received sucralfate. Clinically important upper GI bleeding occurred in 1.7% and 8.1% of participants in the two groups. Data seemed to show a beneficial effect of H2 receptor antagonists plus antacids compared with sucralfate with respect to the occurrence of upper GI bleeding (RR 0.24, 95% CI 0.06 to 0.95). We judged the certainty of this evidence as moderate.

Nosocomial pneumonia

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 15.2). In the three included studies, 144 participants received cimetidine or ranitidine plus antacids, and 137 received sucralfate. Nosocomial pneumonia occurred in 25.0% and 24.1% of participants in the two groups. For this outcome, heterogeneity was substantial (Chi² = 6.36, df = 2.0 (P = 0.04), I² = 69%), and data seemed to show no difference between the two groups with respect to the occurrence of nosocomial pneumonia (RR 1.09, 95% CI 0.51 to 2.32). We judged the certainty of this evidence as very low.

All‐cause mortality in ICU

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 15.3). In the two included studies, 119 participants received cimetidine or ranitidine plus antacids, and 111 received sucralfate. All‐cause mortality occurred in 35.3% and 25.2% of participants in the two groups. Data seemed to show no difference between the two groups with respect to all‐cause mortality in the ICU (RR 1.38, 95% CI 0.92 to 2.05).

Duration of ICU stay

We classified this study on the basis of its intervention and comparison arms (as shown in Analysis 15.4). In the included study, 69 participants received cimetidine or ranitidine plus antacids, and 61 received sucralfate. Data seemed to show no difference between the two groups with respect to duration of ICU stay (MD 3.60 days, 95% CI ‐1.11 to 8.31). We judged the certainty of this evidence as low.

Duration of intubation

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 15.5). In the two included studies, 119 participants received cimetidine or ranitidine plus antacids, and 111 received sucralfate. Data seemed to show no difference in duration of intubation for both study groups (MD ‐1.24 days, 95% CI ‐13.82 to 11.33). Heterogeneity was substantial for this outcome (Chi² = 4.32, df = 1.0 (P = 0.04), I² = 77%).

Number of participants requiring blood transfusions

We classified this study on the basis of its intervention and comparison arms (as shown in Analysis 15.6). In the included study, 69 participants received cimetidine or ranitidine plus antacids, and 61 received sucralfate. Blood transfusions were required by 0% and 1.6% of participants in both groups. Data seemed to show no difference between the two groups with respect to the number of participants requiring blood transfusion (RR 0.30, 95% CI 0.01 to 7.12). We judged the certainty of this evidence as low.

Clinically important upper GI bleeding

We classified this study on the basis of its intervention and comparison arms (as shown in Analysis 16.1). In the single included study, 70 participants received proton pump inhibitors and 70 received teprenone. Clinically important upper GI bleeding occurred in 0% and 5.7% of participants in both groups. We found no evidence of a clear difference between the two groups in this comparison (RR 0.11, 95% CI 0.01 to 2.03; Analysis 16.1). We judged the certainty of this evidence as low.

All‐cause mortality in ICU

We classified this study on the basis of its intervention and comparison arms (as shown in Analysis 16.2). In the single included study, 70 participants received proton pump inhibitors and 70 received teprenone. All‐cause mortality occurred in 0% and 0.7% of participants in both groups. Data show no clear difference between proton pump inhibitors and teprenone (RR 0.33, 95% CI 0.01 to 8.04). We judged the certainty of this evidence as moderate.

Number of participants requiring blood transfusion

We classified this study on the basis of its intervention and comparison arms (as shown in Analysis 16.3). In the single included study, 70 participants received proton pump inhibitors and 70 received teprenone. Data show no clear difference between proton pump inhibitors and teprenone with respect to the number of participants requiring blood transfusion (RR 0.11, 95% CI 0.01 to 2.03). We judged the certainty of this evidence as moderate.

Clinically important upper GI bleeding

We classified this study on the basis of its intervention and comparison arms (as shown in Analysis 17.1). Clinically important upper GI bleeding occurred in 1.6% and 8.33% of participants of the two groups, respectively. We found no evidence of a clear difference between the two groups in this comparison (RR 0.19, 95% CI 0.02 to 1.65; one study; 120 participants).

All‐cause mortality in the hospital

We classified this study on the basis of its intervention and comparison arms (as shown in Analysis 17.2). All‐cause mortality occurred in 5.0% and 21.7% of participants in the two groups, respectively. Data seemed to show a beneficial effect of pantoprazole plus naloxone on all‐cause mortality in the hospital (RR 0.23, 95% CI 0.06 to 0.89; one study; 120 participants).

Adverse events

We classified this study on the basis of its intervention and comparison arms (as shown in Analysis 17.3). Gastrointestinal discomfort, the only adverse event reported, occurred in 10.0% and 18.33% of participants in the two groups, respectively. We found no evidence of a clear difference between the two groups for this comparison (RR 0.40, 95% CI 0.14 to 1.14; one study; 120 participants).

Clinically important upper GI bleeding

We classified this study on the basis of its intervention and comparison arms (as shown in Analysis 18.1). Clinically important upper GI bleeding occurred in 0% and 10.0% of participants in the two groups, respectively. We found no evidence of a clear difference between the two groups for this comparison (RR 0.08, 95% CI 0.00 to 1.34).

Nosocomial pneumonia

We classified this study on the basis of its intervention and comparison arms (as shown in Analysis 18.2). Nosocomial pneumonia occurred in 6.7% and 10% of participants in the two groups, respectively. We found no evidence of a clear difference between the two groups in this comparison (RR 0.67, 95% CI 0.20 to 2.24).

All‐cause mortality in hospital

For this outcome, we found a single study (N = 120). All‐cause mortality in the hospital occurred in 10% and 0% of participants in the two groups, respectively. Data seemed to show no clear difference between proton pump inhibitors and other medication (not defined) (RR 5.00, 95% CI 0.25 to 102.00).

Clinically important upper GI bleeding

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 19.1). In the 16 included studies (N = 1772), 890 participants received antacids and 882 received sucralfate. The event occurred in 6.6% of participants in both groups. Data seemed to show no clear differences between the two groups with respect to the occurrence of clinically important upper GI bleeding (RR 1.00, 95% CI 0.72 to 1.39). We judged the certainty of this evidence as low.

Nosocomial pneumonia

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 19.2). In the seven studies that contributed to this outcome, 501 participants received antacids and 495 received sucralfate. The event occurred in 24.4% and 23.2% of participants in the two groups. Data seemed to show no clear difference between the two groups with respect to the occurrence of nosocomial pneumonia (RR 1.04, 95% CI 0.84 to 1.30). We judged the certainty of this evidence as low.

All‐cause mortality in ICU

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 19.3). In the 11 studies that contributed to this outcome (N = 1249), 628 participants received antacids and 621 received sucralfate. The event occurred in 23.9% and 20.6% of participants in the two groups. Data seemed to show no clear difference between the two groups with respect to all‐cause mortality in the ICU (RR 1.15, 95% CI 0.93 to 1.40). We judged the certainty of this evidence as low.

All‐cause mortality in hospital

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 19.4). In the three studies that contributed to this outcome (N = 450), 227 participants received antacids and 223 received sucralfate. The event occurred in 33.0% and 33.2% of participants in both groups. Data seemed to show no clear difference between the two groups with respect to all‐cause mortality in the hospital (RR 0.98, 95% CI 0.69 to 1.39). This outcome had moderate heterogeneity (Chi² = 3.38, df = 2.0 (P = 0.18), I² = 41%).

Duration of ICU stay

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 19.5). In the two studies that contributed to this outcome, 120 participants received antacids and 107 received sucralfate. Data seemed to show no clear difference between the two groups with respect to duration of ICU stay (MD ‐2.50 days, 95% CI ‐6.61 to 1.61). We rated the certainty of this evidence as low.

Duration of intubation

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 19.6). In the four studies that contributed to this outcome (N = 281), 142 participants received antacids and 139 received sucralfate. Data seemed to show no clear difference between the two groups with respect to duration of intubation (standardised mean difference (SMD) ‐0.18, 95% CI ‐0.41 to 0.06).

Number of participants requiring blood transfusion

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 19.7). In the six studies that contributed to this outcome (N = 667), 338 participants received antacids and 329 received sucralfate. The event occurred in 3.6% and 5.2% of participants in the two groups. Data seemed to show no clear difference between the two groups with respect to the number of participants requiring blood transfusion (RR 0.73, 95% CI 0.40 to 1.34). We judged the certainty of this evidence as low.

Adverse events of interventions

We classified studies on the basis of their intervention and comparison arms and adverse events of the intervention (as shown in Analysis 19.8). We identified nine studies that were relevant to this outcome and categorised data into six adverse events as reported in these studies.

Clinically important upper GI bleeding

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 20.1). In the two included studies (N = 329), 162 participants received antacids and 185 participants received prostaglandin analogues. This event occurred in 2.4% and 7.8% of participants in the two groups. Data seemed to show a beneficial effect of antacids compared with prostaglandin analogues on the occurrence of clinically important upper GI bleeding (RR 0.33, 95% CI 0.12 to 0.91). Results favour antacids. We judged the certainty of this evidence as moderate.

All‐cause mortality in ICU

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 20.2). In the two studies that contributed to this outcome (N = 417), 206 participants received antacids and 211 received prostaglandin analogues. This event occurred in 6.3% and 7.6% of participants in both groups. Data show no difference between the two groups with respect to all‐cause mortality in the ICU (RR 0.84, 95% CI 0.42 to 1.67). We judged the certainty of this evidence as low.

Adverse events of interventions

We classified this study on the basis of its intervention and comparison arms and adverse events of the intervention (as shown in Analysis 20.3).

Clinically important upper GI bleeding

We classified this study on the basis of its intervention and comparison arms (as shown in Analysis 21.1). Clinically important upper GI bleeding occurred in 5.3% of participants in the antacid arm and in 8.2% of participants in the bioflavonoid arm. Data show no difference between the two groups with respect to the occurrence of clinically important upper GI bleeding (RR 0.64, 95% CI 0.22 to 1.85; 198 participants; one study). We judged the certainty of this evidence as low.

Number of participants requiring blood transfusion

We classified this study on the basis of its intervention and comparison arms (as shown in Analysis 21.2). In the single included study, 113 participants received antacids and 85 received bioflavonoids. This event occurred in 2.4% of participants in the bioflavonoid arm and in no participants in the antacids arm. Data show no difference between the two groups with respect to the number of participants requiring blood transfusion (RR 0.15, 95% CI 0.01 to 3.10; 198 participants; one study). We judged the certainty of this evidence as low.

Clinically important upper GI bleeding

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 22.1). In the three included studies (N = 287), 139 participants received sucralfate and 148 received proton pump inhibitors (omeprazole). Clinically important upper GI bleeding occurred in 5.8% and 2% of participants in both groups. Data showed no difference between the two groups with respect to the occurrence of clinically important upper GI bleeding (RR 2.58, 95% CI 0.77 to 8.63). We judged the certainty of this evidence as low.

Ventilator‐associated pneumonia

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 22.2). In the four studies that contributed to this outcome (N = 424), 210 participants received sucralfate and 214 received proton pump inhibitors. Nosocomial pneumonia occurred in 16.6% and 25.2% of participants in the two groups. Data showed no difference between the two groups with respect to the occurrence of nosocomial pneumonia (RR 0.67, 95% CI 0.41 to 1.09). We judged the certainty of this evidence as low.

All‐cause mortality in ICU

We classified studies n the basis of their intervention and comparison arms (as shown in Analysis 22.3). In the four studies that contributed to this outcome (N = 424), 205 participants received sucralfate and 219 received proton pump inhibitors (omeprazole). All‐cause mortality in the ICU occurred in 15.6% and 14.6% of participants in both groups. Data show no significant differences between the two groups with respect to all‐cause mortality in the ICU (RR 1.07, 95% CI 0.68 to 1.68). We judged the certainty of this evidence as low.

All‐cause mortality in hospital

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 22.4). In the two studies that contributed to this outcome (N = 278), 140 participants received sucralfate and 138 received proton pump inhibitors. This event occurred in 13.5% and 17.4% of participants in the two groups. Data showed no significant difference between the two groups with respect to all‐cause mortality in the hospital (RR 0.79, 95% CI 0.46 to 1.37).

Duration of ICU stay

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 22.5). In the two studies that contributed to this outcome (N = 217), 107 participants received sucralfate and 110 received proton pump inhibitors. Data seemed to show no difference between the two groups with respect to duration of ICU stay (MD 0.01 days, 95% CI ‐1.68 to 1.70). We judged the certainty of this evidence as low.

Duration of intubation

We classified studies on the basis of their intervention and comparison arms (as shown in Analysis 22.6). In the three studies that contributed to this outcome (N = 354), 173 participants received sucralfate and 181 received proton pump inhibitors. Data seemed to show no significant difference between the two groups with respect to duration of intubation (MD ‐0.16 days, 95% CI ‐1.61 to 1.28).

Number of participants requiring blood transfusion

We classified this studies on the basis of their intervention and comparison arms (as shown in Analysis 22.7). In the single study that contributed to this outcome (N = 70), 32 participants received sucralfate and 38 received proton pump inhibitors. Blood transfusions were required by 6.25% and 0% of participants in the two groups, respectively. Data seemed to show differences between the two groups with respect to the number of participants requiring blood transfusion (RR 5.91, 95% CI 0.29 to 118.78). We judged the certainty of this evidence as moderate.

Adverse events of interventions

For this outcome, we found a single study and categorised data according to eight adverse events (N = 1096).

Clinically important upper GI bleeding

We classified this study on the basis of its intervention and comparison arms (as shown in Analysis 23.1). In the single included study (N = 185), 100 participants received sucralfate and 85 received bioflavonoids. Clinically important upper GI bleeding occurred in 9% of participants in the antacid arm and in 8.2% of participants in the bioflavonoid arm. Data showed no clear difference between the two groups with respect to the occurrence of clinically important upper GI bleeding (RR 1.09, 95% CI 0.43 to 2.81). We judged the certainty of this evidence as low.

Number of participants requiring blood transfusions

We classified this study on the basis of its intervention and comparison arms (as shown in Analysis 23.2). In the single included study (N = 185), 100 participants received antacids and 85 received bioflavonoids. Blood transfusions were required by 1% of participants in the sucralfate arm and by 2.4% of participants in the bioflavonoid arm. Data showed no clear difference between the two groups with respect to the number of participants requiring blood transfusion (RR 0.42, 95% CI 0.04 to 4.61). Results did not favour any intervention. We judged the certainty of this evidence as low.

Clinically important upper GI bleeding

For this outcome, we found a single study and categorised data by two comparisons. We judged the certainty of this evidence as low.

All‐cause mortality in ICU

For this outcome, we found a single study and categorised data by two comparisons. We judged the certainty of this evidence as low.

Duration of intubation

For this outcome, we found a single study and categorised data by two comparisons. We judged the certainty of this evidence as low.

Clinically important upper GI bleeding

We classified this study on the basis of its intervention and comparison arms (as shown in Analysis 25.1). Clinically important upper GI bleeding occurred in 14.0% and 36.0% of participants in the two groups, respectively. We found evidence of a clear difference between bowel stimulation and no prophylaxis (RR 0.39, 95% CI 0.18 to 0.85). Results favour the bowel stimulation protocol arm.

Clinically important upper GI bleeding

We classified this study on the basis of its intervention and comparison arms (as shown in Analysis 26.1). Clinically important upper GI bleeding occurred in 15.4% and 5.6% of participants in the two groups, respectively. For this outcome, we found evidence suggesting that nasojejunal nutrition was clearly different in its effect compared with nasogastric nutrition (RR 2.74, 95% CI 1.03 to 7.28). Results favour nasogastric nutrition. We judged the certainty of this evidence as moderate.

Nosocomial pneumonia

We classified this study on the basis of its intervention and comparison arms (as shown in Analysis 26.2). Nosocomial pneumonia occurred in 19.8% and 21.3% of participants in the two groups. Data seemed to show no difference between nasojejunal nutrition and nasogastric nutrition (RR 0.93, 95% CI 0.52 to 1.65). We judged the certainty of this evidence as low.

All‐cause mortality in hospital

We classified this study on the basis of its intervention and comparison arms (as shown in Analysis 26.3). All‐cause mortality occurred in 14.3% and 13.5% of participants in the two groups. Data seemed to show no clear difference between nasojejunal nutrition and nasogastric nutrition (RR 1.06, 95% CI 0.51 to 2.19).

Duration of ICU stay

This study did not report enough data for inclusion in formal meta‐analyses. The median duration of the ICU stay was 11 days and 10 days in the nasogastric group (n = 89) and the nasojejunal group (n = 91), respectively.

Duration of mechanical ventilation

This study did not report enough data for inclusion in formal meta‐analyses. The median duration of mechanical ventilation was eight days in both the nasogastric group (n = 89) and the nasojejunal group (n = 91).

Adverse events of interventions

We classified this study on the basis of its intervention and comparison arms (as shown in Analysis 26.4). Researchers reported 77 and 76 adverse events, including vomiting, witnessed aspiration, abdominal distension, and diarrhoea, in the two groups, respectively. We found no evidence of a clear difference between the two groups in this comparison (RR 0.99, 95% CI 0.88 to 1.12).

Nosocomial pneumonia

We classified this study on the basis of its intervention and comparison arms (as shown in Analysis 27.1). Nosocomial pneumonia occurred in 27.5% and 35.0% of participants in the two groups, respectively. Data seemed to show no clear difference between enteral plus parenteral nutrition and interventions in the other study groups (RR 0.79, 95% CI 0.44 to 1.40; one study; 120 participants).

All‐cause mortality in the hospital

We classified this study on the basis of its intervention and comparison arms (as shown in Analysis 27.2). All‐cause mortality in the hospital occurred in 10.0% and 36.2% of participants in the two groups, respectively. All‐cause mortality seemed to be lower in the group receiving enteral plus parenteral nutrition than in the groups receiving other nutrition regimens (RR 0.20, 95% CI 0.06 to 0.60; one study; 120 participants).

Duration of ICU stay

We classified this study on the basis of its intervention and comparison arms (as shown in Analysis 27.3). Duration of ICU stay seemed to be shorter in the study group receiving enteral plus parenteral nutrition compared with groups receiving other nutrition regimens (MD ‐5.98 days, 95% CI ‐8.81 to ‐3.16; one study; 120 participants).

Duration of intubation

We classified this study on the basis of its intervention and comparison arms (as shown in Analysis 27.4). Duration of intubation seemed to be shorter in the study group receiving enteral plus parenteral nutrition compared with the groups receiving other nutrition regimens (MD ‐7.37 days, 95% CI ‐9.29 to ‐5.45; one study; 120 participants).

Adverse events of interventions

We classified this study on the basis of its intervention and comparison arms (as shown in Analysis 27.5 and following).

Presence of bleeding disorders

We identified three studies that explicitly reported including patients with bleeding disorders (Conrad 2005; Kantorova 2004; Martin 1992). Only two studies compared a common intervention and a comparator (H2 receptor antagonists vs proton pump inhibitors). Subgroup analyses provided no evidence of a difference between studies that explicitly included patients with bleeding disorders and those that did not include such patients (test for subgroup differences P value = 0.09). Further, there seemed to be no difference between groups with respect to the occurrence of nosocomial pneumonia (test for subgroup differences P value = 0.37) and all‐cause mortality in the ICU (test for subgroup differences P value = 0.52). Researchers provided no data on other outcomes and comparisons for this subgroup that could be formally included in meta‐analysis.

Pneumonia at time of admission

We identified four studies that explicitly reported including participants with pneumonia at the time of ICU admission (Ben‐Menachem 1994; Karlstadt 1990; Martin 1993; Metz 1993). The proportion of participants with pneumonia ranged from 2.33% to 21% in these studies, respectively. All studies compared the effects of H2 receptor antagonists (cimetidine or ranitidine) versus placebo or no prophylaxis. Subgroup analyses provided no evidence of a difference between studies that explicitly included patients with pneumonia at the time of admission and those that did not include such patients (test for subgroup differences P value = 0.28). Data showed no difference between groups with respect to the occurrence of nosocomial pneumonia (test for subgroup differences P value = 0.54) and all‐cause mortality in the ICU (test for subgroup differences P value = 0.20). Researchers provided no data on other outcomes and comparisons for this subgroup that could be formally analysed in meta‐analysis.

Infants and children

We identified five studies that explicitly reported including only children and adolescents (Behrens 1994; Kuusela 1997; Lacroix 1986; Lopez‐Herce 1992; Yildizdas 2002). All participants in these studies were aged from 0 to 20 years. Two studies compared effects of H2 receptor antagonists versus placebo, and three studies had multiple study arms including H2 receptor antagonists, anticholinergics (Behrens 1994), antacids (Lopez‐Herce 1992), sucralfate (Lopez‐Herce 1992;Yildizdas 2002), and proton pump inhibitors (Yildizdas 2002), as well as placebo or no prophylaxis. In infants and adolescents, data showed no difference between H2 receptor antagonists and placebo or no prophylaxis with respect to the occurrence of clinically important upper GI bleeding (test for subgroup differences P value = 0.10), nosocomial pneumonia (test for subgroup differences P value = 0.84), or all‐cause mortality in the ICU (test for subgroup differences P value = 0.52); duration of ICU stay (test for subgroup differences P value = 0.62); duration of intubation (test for subgroup differences P value = 0.75); or number of participants requiring blood transfusion (test for subgroup differences P value = 0.37). Researchers provided no data on other outcomes of this comparison for this subgroup that could be formally included in meta‐analysis.

We detected no differences between H2 receptor antagonists and sucralfate with respect to the occurrence of clinically important upper GI bleeding in infants (test for subgroup differences P value = 0.61), nosocomial pneumonia (test for subgroup differences P value = 0.75), all‐cause mortality in the ICU (test for subgroup differences P value = 0.92), duration of ICU stay (test for subgroup differences P value = 0.84), or duration of intubation (test for subgroup differences P value = 0.90). We found no data on other outcomes in this comparison for this subgroup that could be formally analysed in meta‐analysis. Moreover, researchers provided no data on other outcomes and comparisons for this subgroup that could be formally analysed in meta‐analysis.

The reporting quality of the included studies did not permit other subgroup analyses. No studies explicitly included only adults over the age of 65 years. Thus, insufficient data were available for analysis of the effect of bleeding prophylaxis in older adults only. Furthermore, use of co‐interventions and the level of detail of reporting about co‐interventions varied strongly among studies, so that meaningful subgroup analyses were not possible with respect to differences in effect due to co‐interventions.

Studies with low risk of bias versus studies with unclear or high risk of bias

Only two studies had an overall low risk of bias (Cook 1998; Ephgrave 1998), and one study had low risk of bias, when risk of performance and detection bias was assessed for upper GI bleeding only (Ng 2012).

Cook 1998 compared H2 receptor antagonists (ranitidine) (n = 596) versus sucralfate (n = 604). This study found lower risk of upper GI bleeding with H2 receptor antagonists (RR 0.44, 95% CI 0.21 to 0.92; one study; 1200 participants) compared with studies with unclear and high risk of bias (RR 1.28, 95% CI 0.98 to 1.66; 23 studies; 2199 participants). Data seemed to show no difference between study groups with respect to the occurrence of nosocomial pneumonia in the study with low risk of bias (RR 1.18, 95% CI 0.92 to 1.51; one study; 1200 participants), and studies with unclear or high risk of bias found increased risk of nosocomial pneumonia (RR 1.21, 95% CI 1.02 to 1.43; 14 studies; 1807 participants). The latter result favoured the sucralfate arm. Data show no differences between studies with low risk of bias and studies with unclear or high risk of bias in the comparison of H2 receptor antagonists versus sucralfate with respect to the outcome all‐cause mortality in the ICU, duration of intubation, and duration of ICU stay. No studies reported data on other outcomes pre‐defined in this review, so that no further sensitivity analyses were possible.

Ephgrave 1998 compared antacids (n = 70) versus sucralfate (n = 70) and reported no differences in effect between the study with low risk of bias and studies with unclear or high risk of bias with respect to the outcomes clinically important upper GI bleeding, nosocomial pneumonia, and all‐cause mortality in the ICU. No other outcomes of this comparison were available for sensitivity analysis.

Ng 2012 compared an H2 receptor antagonist (famotidine) (n = 149) versus a proton pump inhibitor (esomeprazole) (n = 164) and reported no differences in effect between the study with low risk of bias and studies with unclear or high risk of bias with respect to the outcomes clinically important upper GI bleeding and all‐cause mortality in the hospital. No studies reported data on other outcome pre‐defined in this review, so that no further sensitivity analyses were possible.

Studies with attrition greater than 10%

Six studies had an attrition rate greater than 10% (Barandun 1985;Fabian 1993;Hanisch 1998;Israsena 1987;Ruiz‐Santana 1991;Terzi 2009).

Barandun 1985 compared H2 receptor antagonists (n = 28) versus anticholinergics (n = 27) and noted no apparent differences in this comparison with respect to the occurrence of clinically important upper GI bleeding when excluding from analysis the study with an attrition rate of 10% or higher. No other outcomes of this comparison were available for sensitivity analysis.

Fabian 1993 compared H2 receptor antagonists (n = 410) versus sucralfate (n = 206) and noted no apparent differences in this comparison with respect to the occurrence of clinically important upper GI bleeding, nosocomial pneumonia, all‐cause mortality in the ICU or the hospital, duration of ICU stay, and the number of participants requiring blood transfusions when excluding from analysis the study with an attrition rate of 10% or higher. No other outcomes of this comparison were available for sensitivity analysis.

Hanisch 1998 compared an H2 receptor antagonist (ranitidine) (n = 57) versus an anticholinergic (pirenzepine) (n = 44) or placebo (n = 57) and noted no apparent difference in effect size and direction when excluding the study with an attrition rate of 10% from the comparison of H2 receptor antagonists versus anticholinergics or placebo. Neither did direction or strength of effect change when the study with an attrition rate of 10% was removed from the comparison of anticholinergics and placebo.

Israsena 1987 compared antacids (n = 47) versus sucralfate (n = 40) and noted no apparent difference in effect size and direction when excluding from analysis the study with an attrition rate of 10%.

Ruiz‐Santana 1991 was the only study comparing total parenteral nutrition with total parenteral nutrition plus other medications. So, no sensitivity analysis could be performed.

Terzi 2009 compared an H2 receptor antagonist (ranitidine) (n = 10) verus a proton pump inhibitor (pantoprazole) (n = 10) and noted no apparent difference in effect size and direction when excluding from analysis the study with an attrition rate of 10%.

Studies with published and validated criteria to diagnose clinically important upper GI bleeding and nosocomial pneumonia

Lack of detail regarding criteria used to diagnose the primary endpoints of the review did not allow sensitivity analysis.