Twenty RCTs were included (1,971 intensive care unit patients, range 25 to 208). Length of follow up ranged from seven to 29 months. All of the trials were judged to be at high risk of bias. Few trials reported adequate randomisation methods, allocation concealment or blinding.
There was a significant difference in gastrointestinal bleeding in favour of stress ulcer prophylaxis (random-effects RR 0.44, 95% CI 0.28 to 0.68; 20 RCTs; Ι²=48%; fixed-effect results similar). This beneficial effect was not found in any of the subgroup analyses or trial sequential analysis.
There were no significant differences in all-cause mortality (fixed-effect RR 1.00, 95% CI 0.84 to 1.20; 15 RCTs; Ι²=0%) or hospital-acquired pneumonia (random-effects RR 1.23, 95% CI 0.86 to 1.78; seven RCTs; Ι²=19%; fixed-effect results similar) between patients treated with stress ulcer prophylaxis and those treated with placebo or no prophylaxis.
Results of subgroup analyses and trial sequential analysis did not differ for these two outcomes. Sensitivity analysis did not change the results for any outcome. Trial sequential analysis showed that there were insufficient numbers of patients included to detect a clinically meaningful difference in the meta-analysis for all three outcomes.
There was possible publication bias for the all-cause mortality outcome.