The PRISMA flowchart of this study is shown in Figure 1 and the reasons for exclusion of illegible studies are presented in Table S1 in the supplementary materials. We searched a total of 10562 records in the abovementioned database, additional 119 records from other meta-analyses were also identified. Among these records, 946 duplicate records and 9630 records, which were ineligible for our inclusion criteria, were excluded after screening the title and abstract. Then, the remaining 105 studies were included in the review of the full text. Finally, a total of 29 studies [7–13,18, 24–44] met the eligibility criteria and were included in this meta-analysis.
Characteristics of included studies
The detailed characteristics and clinical outcomes of the individual studies are described in the supplementary materials (Table S2 and S3). Among these included RCTs [7–13,18,24–44], six trials [18,24,30,36,38,43] were published in recent 3 years, seven trials [11–13,18,24,30,44] were multi-center, and twenty-two trials [7–10,25–29,31–43] were single-center. The discrepancy in the number of subjects between trials were large and it varied from 25 to 3298. The subjects were from medical ICU [7,25,26,30,31,38,40], surgical ICU [8,10,13,28,34,35,37,39,41,44], or both [9,11,12,18,24,27,29,32,33,36,42,43]. SUP with H2RA was reported in 20 trials [7–13,25–29,31,32,34,38,40–42,44], of which 9 trials [9,13,25,28,29,34,38,41,42] used ranitidine and 11 trials [7,8,10–12,26,27,31,32,40,44] used cimetidine. There are 6 trials in which the intervention group received PPI as SUP [8,24,30,33,36,43], and it is worth mentioning that the trial by Gursoy et al.  set 5 intervention groups and one control group, all intervention groups received PPI, hence we combined the intervention groups into a single group to compare with the control group. Notably, the trials by Kantorova et al.  and Liu et al.  set 2 intervention groups in which one group used PPI and another group used H2RA, therefore the control group was divided evenly into two groups to make two comparisons. Additionally, as to the trial by Powell et al.  which have 3 intervention groups (one group received H2RA and two groups received PPI), we combined the two groups received PPI into a single group, and divided the control group into two groups. In all included trials, the dosage, duration of drug used and drug-delivery methods were largely different in the intervention group. The control groups received placebo[10–13,18,24,27–35,37,39,40,43] or no prophylaxis [7–9,25,26,36,38,41,42,44]. The subjects in 16 trials did not receive EN [7–9,11–13,28,29,31–34,37,39,41,44], however part or all patients received EN in the remaining 13 trials [10,18,24–27,30,35,36,38,40,42,43].
Risk of bias assessment
The full details of the risk of bias assessment of each included trials were presented in Table 2. There were 4 studies adjudicated as low risk of bias in all domains [18, 24,30,34], and all other studies were deemed overall high risk of bias due to at least one domain with an unclear or high risk of bias.
Analysis of the primary outcomes
The clinically important GI bleeding
There were 11 studies[10,11,18,24,26,28,30,34–36,43], consisting of 4521 participants, reported data on the clinically important GI bleeding. The pooled results demonstrated that use of SUP was associated with a decreased risk of the clinically important GI bleeding (RR = 0.58, 95% CI: 0.42–0.81, P = 0.001,I2 = 0%), this benefit was also found in sub-analysis of trials with low risk of bias (RR = 0.64, 95% CI: 0.45–0.92, P = 0.017;I2 = 0%) (Figure 2.A), trials used PPI (RR = 0.61, 95% CI: 0.43–0.88, P = 0.008; I2 = 0%) (Figure 2.B), and trials received EN (RR = 0.61, 95% CI: 0.44–0.85, P = 0.004; I2 = 0%) (Figure 2.C), but not in trials used H2RA and trials did not receive EN. However, the results from TSA showed that the Z-curve just crossed the conventional boundary for benefit, but not the trial sequential monitoring boundary for benefit (Figure 3 and Figure S1, in the supplementary marterial), indicated that it still lack of sufficient evidence to favor a 20% relative risk reduction in the risk of clinically important GI bleeding with use of SUP.
The overt GI bleeding
There were 27 studies[7–13,18,24–28,30–32, 34–44], with 30 comparisons, including 5919 participants, reported available data regarding overt GI bleeding. The pooled RR from the conventional meta-analysis for overt GI bleeding in all trials was 0.48 (95% CI: 0.36–0.63, P<0.001;I2 = 34.6%), the TSA showed that the trial sequential monitoring boundary for benefit was crossed with TSA-adjusted CI from 0.31 to 0.75 (D2 = 62%) (Figure 4), indicating a firm evidence on the beneficial effect of SUP on overt GI bleeding. In the sub-analyses of trials based on the trials quality (Figure 5.A), the type of SUP used (Figure 5.B), and whether EN was used (Figure 5.C), we also found a significant reduction in the incident of overt GI bleeding with use of SUP. Furthermore, TSA showed that the Z-curve crossed the trial sequential monitoring boundary for benefit in trials did not receive EN (Figure 6.A), in trials used PPI as SUP (Figure S2.A), and in trials used H2RA as SUP (Figure S2.B). However, the TSA for trials adjudicated as low risk of bias (Figure 6.B) and trials received EN (Figure 6.C) revealed a neutral benefit in reducing the risk of overt GI bleeding, thus more evidence is required to confirm this benefits in this case.
Analysis of the secondary outcomes
There are 24 RCTs [7,10–12,18,24–26,28–30,32–44] included in analysis for all-cause mortality, the conventional meta-analysis indicated that SUP was not associated with a lower mortality compared with the control group, regardless of the trials quality (Figure S3.A), the type of SUP used (Figure S3.B), and whether EN was used (Figure S3.C). These results was firmly confirmed by TSA in which the Z-curve reached the futility area, indicating that there are sufficient events in current trials to reject a 20% relative risk reduction in mortality with use of SUP (Figure S4). However, it still lack of sufficient evidence on the effects of SUP on mortality in patients did not receive EN (Figure S4.F), because the Z-curve did not cross any boundaries.
In the meta-analysis of 12 trials [11,12,18,24–26,34–38,43] refered to data on pneumonia, we found that administration of SUP was not accompanied by increased risk of pneumonia, TSA further confirmed this result with the Z-curve crossing the boundary for futility (Figure S5.A). Also, negtive results were observed in sub-analysis of trials with low risk of bias (Figure S6.A), in trials used PPI as SUP(Figure S6.B), in trials used H2RA as SUP (Figure S6.B), in trials with use of EN (Figure S6.C), or in trials without use of EN (Figure S6.C), TSA suggested that the pooled results were robust in subgroup of trials with low risk of bias (Figure S5.B), trials received PPI (Figure S5.C), and trials with use of EN(Figure S5.E).
There are 4 trials presented data on CDI [18,24,30,43], all of them received PPI and EN, three of them was judged have low risk of bias. There was no difference in the incident of CDI between SUP group and control group. Sub-analysis of trials with low risk of bias also revealed a neutral effect of SUP on the incident of CDI (Figure S7). However, TSA demonstrated that the Z-curve crossed no boundaries (Figure S8), indicating that this results could be changed when more studies were incuded.
Analysis of the tertiary outcomes
There are 11 studies reported data on the duration of ICU stay [12,24,26,30,33–35,38,41–43] and 7 studies [24,26,30,34,35,38,43] reported data on the duration of MV, the pooled results suggested that use of SUP could not result in a significant reduction in the duration of ICU stay (Figure S9) and duration of MV (Figure S10) when compared with placebo or no prophylaxis, regardless of the trials quality, the type of SUP used, and whether EN was used.
Analysis for publication bias
Publication bias was evaluated for the primary outcomes. All the funnel plots were visually symmetrical (Figure 7). The Begg’s and Egger’s tests for the clinically important GI bleeding (P = 0.436 and P = 0.395, respectively) and for overt GI bleeding (P = 0.617 and P = 0.280, respectively) revealed no evidence of publication bias.