Fourteen trials (1,720 patients) were included. Three trials were judged to be at low risk of bias, six trials had high risk of bias and five had an unclear risk of bias. Most of the high risk was due to inadequate blinding (though further details on this were not presented). Four trials were published in abstract form only.
Proton pump inhibitors appeared more effective than histamine 2 receptor antagonists at reducing clinically important upper gastrointestinal bleeding (RR 0.36, 95% CI 0.19 to 0.68; 12 RCTs, five without any events; Ι²=0%) and overt upper gastrointestinal bleeding (RR 0.35; 95% CI 0.21 to 0.59, 14 RCTs, five without any events; Ι²=15%). There was evidence of publication bias for the clinically important bleeding outcome.
For both of the above outcomes, trials at low risk of bias were significantly associated with a smaller treatment effect when compared with trials having a high or unclear risk of bias; when results from only the low risk of bias studies were pooled there were no statistically significant differences between treatments for both outcomes. When absolute risk difference was used as an effect estimate (in a sensitivity analysis), statistically significant heterogeneity was evident. There were no subgroup differences when investigating the effect of: route of proton pump inhibitor administration, frequency of proton pump inhibitor dosing, intensive care unit type, or trial setting (Asian versus non-Asian).
There were no differences in the risk of nosocomial pneumonia (eight RCTs), intensive care unit mortality (eight RCTs), or intensive care unit length of stay (five RCTs). No trials reported on C. difficile infection.