Probiotics for Preventing Ventilator-Associated Pneumonia in Mechanically Ventilated Patients: A Meta-Analysis with Trial Sequential Analysis

Background and Objective: Ventilator-associated pneumonia (VAP) is still an important cause of morbidity and mortality in mechanically ventilated patients. The efficacy of the probiotics for preventing VAP is still controversial. Present study was conducted to comprehensively evaluate the effect of probiotics on VAP prevention in mechanically ventilated patients. Methods: PubMed, Embase, and CENTRAL were searched up to September 2016. Eligible trials designed with randomized controlled trials (RCTs) comparing probiotics with control in mechanically ventilated patients were included. Risk ratios (RRs) and mean differences (MDs) with 95% confidence intervals (CIs) were estimated with fixed or random effects models. Trial sequential analysis (TSA) was performed using TSA 0.9beta software. Results: Thirteen RCTs (N = 1969) were included. Overall, probiotics were associated with reduced incidence of VAP (RR = 0.73, 95% CI = 0.60–0.89; P = 0.002), which was confirmed by TSA (TSA adjusted 95% CI = 0.55–0.96). However, no significant difference was observed in 90-day mortality (RR = 1.00, 95% CI = 0.72–1.37; P = 0.99), overall mortality (RR = 0.84, 95% CI = 0.70–1.02; P = 0.09), 28-day mortality (RR = 1.06, 95% CI = 0.72–1.57; P = 0.99), intensive care unit (ICU) mortality (RR = 0.97, 95% CI = 0.74–1.27; P = 0.82), hospital mortality (RR = 0.81, 95% CI = 0.65–1.02; P = 0.07), diarrhea (RR = 0.99, 95% CI = 0.83–1.19; P = 0.92), length of ICU stay (MD = −2.40 days, 95% CI = −6.75 to 1.95; P = 0.28), length of hospital stay (MD = −1.34 days, 95% CI = −6.21 to 3.54; P = 0.59), and duration of mechanical ventilation (MD = −3.32 days, 95% CI = −6.74 to 0.09; P = 0.06). Conclusions: In this meta-analysis, we found that probiotics could reduce the incidence of VAP in mechanically ventilated patients. It seems likely that probiotics provide clinical benefits for mechanically ventilated patients.


INTRODUCTION
Ventilator-associated pneumonia (VAP) is still an important cause of morbidity and mortality in mechanically ventilated patients even though the incidence thereof has been decreased in the past several years in America (Metersky et al., 2016). It is estimated that VAP may be responsible for ∼27-47% of intensive care unit (ICU) acquired infections (Grap et al., 2012). The clinical and economic burden of VAP remains high and the application of existing VAP prevention strategies is variable but disappointing (Muscedere et al., 2008;Amin, 2009;Kallet, 2015). Therefore, a simple, inexpensive, and safe prevention strategy will contribute to the decrease of VAP occurrence rate and corresponding burden. The pathogenesis of VAP is complicated; however it typically involves the colonization of upper aerodigestive tract with pathogenic bacteria and the leakage of contaminated oropharyngeal secretions into the lung (Kollef, 2005;Baselski and Klutts, 2013). Numerous studies have assessed various strategies of VAP prevention which can be classified into pharmacologic and non-pharmacologic interventions. Compared to other strategies, probiotics have been considered as a new intervention for VAP prevention in critical care medicine.
In recent years, several studies suggest that orally administered probiotics may conduce to the prevention of VAP (Siempos et al., 2010;Theodorakopoulou et al., 2013). However, the conclusions on this topic are still controversial (Siempos et al., 2010;Gu et al., 2012;Wang et al., 2013;Bo et al., 2014). In 2010, Siempos et al. (2010) performed a metaanalysis with five trials and supported that probiotics were associated with decreased risk of VAP, which was further confirmed by a Cochrane systematic review with eight trials (Bo et al., 2014). However, another meta-analysis carried out by Gu et al. (2012) with seven trials concluded that probiotics were not beneficial to mechanically ventilated patients. Additionally, the results of a subsequent meta-analysis performed by Wang et al. (2013) with five trials also demonstrated that probiotics had no beneficial effect for prevention of VAP. Several trials have been applied to assess the role of probiotics in VAP prevention since the previous meta-analyses were published. Additionally, due to uncertain efficacy and safety of probiotics, most ICU pharmacists would not currently recommend this strategy for prevention of VAP (Wheeler et al., 2016). Therefore, we performed an updated meta-analysis to evaluate the effectiveness and safety of probiotics for preventing VAP, thereby providing a more precise evidence for clinical practice.

Eligibility Criteria
This meta-analysis is reported based on the methodology of Cochrane Handbook (Higgins and Green, 2011) and conducted in adherence to Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement (Moher et al., 2009). The inclusion criteria were a s following: (1) patients: the study subjects were mechanically ventilated patients; (2) intervention: probiotics; (3) comparison: placebo or other drugs; (4) outcomes: primary outcome was incidence of VAP; secondary outcomes were 90-day mortality, overall mortality, 28-day mortality, ICU mortality, and hospital mortality; tertiary outcomes were diarrhea, length of ICU stay, length of hospital stay, and duration of mechanical ventilation; (5) study type: only randomized controlled trials (RCTs) that were peer-reviewed and available in full-text would be included in this metaanalysis.

Search Strategy
PubMed, Embase, and CENTRAL on the Cochrane Library were comprehensively searched for all relevant RCTs up to September 2016 by two authors (HW and JL). The following items were combined and adopted to retrieve original studies: "probiotic, " "probiotics, " "prebiotic, " "prebiotics, " "symbiotic, " "synbiotics, " "lactobacillus, " "lactobacilli, " "bifidobacterium, " "pneumonia, " "random, " "placebo, " and "trial." Reference lists of relevant reviews or meta-analyses were manually searched. No language restriction was applied. Any discrepancy was solved by consensus or discussion with a third author (XZ) when necessary.

Data Extraction and Risk of Bias Assessment
Two reviewers (HW and JGL) independently extracted data from eligible studies using a pre-specified data extraction form and assessed the risk of bias of included studies. The extracted information: included name of first author, year of publication, country, institutions, language, funding source, characteristic of participants, details of intervention and comparison treatment, definition of VAP, outcomes, and methodological design. Discrepancy was solved by negotiation between them. The risk of bias of included studies was assessed according to Cochrane Handbook for Systematic Reviews of Interventions criteria (Higgins and Green, 2011).

Statistical Analysis
Dichotomous outcome variables were measured using risk ratios (RRs) and corresponding 95% confidence intervals (CIs). Continuous outcome variables were measured using mean differences (MDs) and corresponding 95% CIs. Heterogeneity between studies was detected by Cochrane's Q-test with P < 0.1 as a significance level, and quantitatively measured through I 2 statistic. Fixed effects model was applied to perform the meta-analysis if the P-value of Cochrane's Q-tests was more than 0.1, otherwise, random effects model was utilized. The statistical significance level was set at 0.05 for this metaanalysis. All the data syntheses were accomplished using RevMan 5.3 software. The number needed to treat (NNT) was also estimated for primary outcome. Sensitivity analyses were performed by excluding studies which would confound the results.
Cumulative meta-analyses of RCTs are at risk of yielding random errors due to sparse data and repetitive testing of accumulating data (Wetterslev et al., 2017). Trial sequential analysis (TSA) depends on the quantification of the required information size (RIS), i.e., optimal information size. TSA was undertaken using TSA 0.9 beta software if the number of included trials was more than five. The RIS was estimated using relative risk reduction and heterogeneity adjusted information size for dichotomous outcomes Wetterslev et al., 2008;Thorlund et al., 2009). The result was confirmed as true positive if the cumulative Z-curve surpassed the Lan-DeMets trial sequential monitoring boundary or reached the RIS above the conventional significance level line (Z = 1.96); and the result was confirmed as true negative if the cumulative Z-curve reached the futility boundary or reached the RIS below the conventional significance level line (Z = 1.96). TSA adjusted 95% CIs were also presented.

Characteristics and Risk of Bias Assessment of Included Trials
We initially retrieved a total of 172 studies from the abovementioned databases. After strict screening according to inclusion criteria, 13 RCTs (Spindler-Vesel et al., 2007;Forestier et al., 2008;Klarin et al., 2008;Giamarellos-Bourboulis et al., 2009;Knight et al., 2009;Barraud et al., 2010;Morrow et al., 2010;Oudhuis et al., 2011;Tan et al., 2011;Li et al., 2012;Banupriya et al., 2015;Rongrungruang et al., 2015;Zeng et al., 2016) were included in the present meta-analysis. The study selection process is presented in Figure 1. Characteristics of included trials  Table 1. These trials were published between 2007 and 2016. The sample sizes of included trials were ranged from 35 to 259 (total number was 1,969). Two studies (Li et al., 2012;Banupriya et al., 2015) focused on children and one study (Klarin et al., 2008) only included probiotics as oral care. These three studies might confound the results of the overall analysis and sensitivity analyses were undertaken by removing these trials for relevant outcomes. Risk of bias assessment of included trials is displayed in Figure 2.

Primary Outcome: Incidence of VAP
The meta-analysis involving 13 trials (1,969 patients) showed a significantly decreased risk in incidence of VAP in patients exposed to probiotics based on random-effects model (RR = 0.73, 95% CI = 0.60-0.89; P = 0.002), as demonstrated in Figure 3. Low to moderate between-study heterogeneity was detected (P = 0.06, I 2 = 40%). The NNT was 10.9 (95% CI = 7.7-19.3). The TSA adjusted 95% CI ranged from 0.55 to 0.96. The TSA result showed that 1,969 (62.9%) of the RIS of 3,132 patients was accrued. The cumulative z-curve crossed the conventional boundary for benefit and crossed the trial sequential monitoring boundary for benefit (Figure 4), indicating that firm evidence of probiotics for preventing VAP was obtained. Sensitivity analysis by removing three trials (Klarin et al., 2008;Li et al., 2012;Banupriya et al., 2015) showed similar results to the overall analysis (RR = 0.86, 95% CI = 0.66-0.97; P = 0.02).

Secondary Outcome 1a: 90-Day Mortality
Two trials concerning 317 patients presented follow-up data up to 90 days. The meta-analysis of these two trials showed no significant difference in 90-day mortality in patients exposed to probiotics based on fixed-effects model (RR = 1.00, 95% CI = 0.72-1.37; P = 0.99), as revealed in Figure 5. No evidence of between-study heterogeneity was detected (P = 0.94, I 2 = 0%).

Secondary Outcome 1b: Overall Mortality
Overall mortality data were obtained from nine RCTs involving 1,296 patients. The meta-analysis of these nine trials indicated no significant difference in overall mortality in patients exposed to probiotics based on fixed-effects model (RR = 0.84, 95% CI = 0.70-1.02; P = 0.09), as shown in Figure 6. No evidence of between-study heterogeneity was detected (P = 0.94, I 2 = 0%). The TSA adjusted 95% CI was ranged from 0.58 to 1.23. The TSA result showed that 1,296 (32.0%) of the RIS of 4,053 patients was accrued. The cumulative z-curve crossed neither the conventional boundary for benefit nor the trial sequential futility boundary for benefit (Figure 7), suggesting that the current evidence was inconclusive. Sensitivity analysis by removing two trials (Klarin et al., 2008;Banupriya et al., 2015) showed similar results to the overall analysis (RR = 0.86, 95% CI = 0.70-1.07; P = 0.17).

Secondary Outcome 1c: 28-Day Mortality
Two trials with 317 patients presented follow-up data up to 28 days. The meta-analysis of these two trials showed no significant difference in 28-day mortality in patients exposed to probiotics based on fixed-effects model (RR = 1.06, 95% CI = 0.72-1.57; P = 0.99), as displayed in Figure 8. No evidence of between-study heterogeneity was detected (P = 0.99, I 2 = 0%).

Secondary Outcome 1d: ICU Mortality
Six trials including 938 patients reported the ICU mortality data. The meta-analysis of these six trials exhibited no significant difference in ICU mortality in patients exposed to probiotics based on fixed-effects model (RR = 0.97, 95% CI = 0.74-1.27; P = 0.82), as shown in Figure 9. No evidence of betweenstudy heterogeneity was detected (P = 0.75, I 2 = 0%). The TSA adjusted 95% CI was ranged from 0.33 to 2.87. The TSA result showed that 938 (15.5%) of the RIS of 6,058 patients was accrued. The cumulative z-curve crossed neither the conventional boundary for benefit nor the trial sequential futility boundary for benefit (Figure 10), revealing that the current evidence was inconclusive. Sensitivity analysis by removing one trial (Klarin et al., 2008) showed similar results to the overall analysis (RR = 0.96, 95% CI = 0.73-1.26; P = 0.78).

Secondary Outcome 1e: Hospital Mortality
Six trials contacting 877 patients reported the ICU mortality data. The meta-analysis of these six trials indicated no significant difference in hospital mortality in patients exposed to probiotics based on fixed-effects model (RR = 0.81, 95% CI = 0.65-1.02; P = 0.07), as shown in Figure 11. No evidence of betweenstudy heterogeneity was detected (P = 0.82, I 2 = 0%). The TSA adjusted 95% CI was ranged from 0.49 to 1.33. The TSA result showed that 877 (25.2%) of the RIS of 3,475 patients was accrued. The cumulative z-curve crossed neither the conventional boundary for benefit nor the trial sequential futility boundary for benefit (Figure 12), revealing that the current evidence was inconclusive. Sensitivity analysis by removing two trials (Klarin et al., 2008;Banupriya et al., 2015) showed similar results to the overall analysis (RR = 0.83, 95% CI = 0.64-1.07; P = 0.15)

Tertiary Outcome 1a: Diarrhea
Five trials with 768 patients reported the diarrhea data. The metaanalysis of these six trials showed no significant difference in diarrhea in patients exposed to probiotics based on fixed-effects model (RR = 0.99, 95% CI = 0.83-1.19; P = 0.92), as presented in Figure 13. No evidence of between-study heterogeneity was detected (P = 0.50, I 2 = 0%).

Tertiary Outcome 1b: Length of ICU Stay
Five trials including 538 patients reported the length of ICU stay. The meta-analysis of these six trials showed no significant difference in length of ICU stay in patients exposed to probiotics based on random-effects model (MD = −2.40 days, 95% CI = −6.75 to 1.95; P = 0.28), as shown in Figure 14. Moderate to high between-study heterogeneity was detected (P = 0.0001, I 2 = 83%). Sensitivity analysis by removing two trials (Klarin et al., 2008;Banupriya et al., 2015) showed similar results to the overall analysis (MD = −3.88 days, 95% CI = −10.51 to 2.76; P = 0.25). One probiotic capsule contained a total of 3.3 billion CFU of probiotic organisms. One capsule was administered twice a day mixed with milk (or 5 ml of 5% dextrose solution if enteral feeding had not been started) and given through a nasogastric tube. A total of 6.6 billion CFU of probiotic organisms per day was administered to each child in the probiotic group for the initial 7 days or till discharge, whichever was earlier Standard care, no placebo VAP was defined as a new (developing more than 48 h after the start of mechanical ventilation or within 48 h of extubation) or persisting radiographic infiltrate (persisting radiographically for at least 72 h) that develops in conjunction with one of the following: (1) Radiographic evidence of pulmonary abscess formation (i.e., cavitations within pre-existing pulmonary infiltrates); (2) Two of the following: fever (increase in the core temperature of at least ) for oral care after the standard oral care once daily. An additional 80 ml of the aforementioned fermented dairy product was given via enteral feeding once daily for 28 days or when their endotracheal tubes were removed. Probiotics was discontinued when diarrhea related to probiotics occurred The patients in the control group did not receive any additional products A diagnosis of VAP was made if the patient had a new, persistent, or progressive infiltrate visible on a chest radiograph in combination with at least 3 of the following 4 criteria: (1) body temperature greater than 38 • C or

Tertiary Outcome 1c: Length of Hospital Stay
Four trials with 682 patients reported the length of hospital stay. The meta-analysis of these six trials showed no significant difference in length of hospital stay in patients exposed to probiotics based on random-effects model (MD = −1.34 days, 95% CI = −6.21 to 3.54; P = 0.59), as displayed in Figure 15.

Tertiary Outcome 1d: Duration of Mechanical Ventilation
Four trials involving 512 patients reported the duration of mechanical ventilation. The meta-analysis of these six trials showed no significant difference in duration of mechanical ventilation in patients exposed to probiotics based on randomeffects model (MD = −3.32 days, 95% CI = −6.74 to 0.09; P = 0.06), as presented in Supplementary Figure 1. Moderate to high between-study heterogeneity was detected (P = 0.0006, I 2 = 83%). Sensitivity analysis by removing one trial (Banupriya et al., 2015) showed similar results to the overall analysis (MD = −3.32 days, 95% CI = −8.03 to 1.38; P = 0.17).

DISCUSSION
To date, the present meta-analysis is the largest and most updated evaluation of the overall effects of probiotics on preventing VAP in mechanically ventilated patients. Based on the analysis of 13 RCTs involving 1,969 patients, we found that probiotics were significantly associated with a decreased risk of VAP in mechanically ventilated patients, which was confirmed by TSA that the result of TSA showed that the cumulative Z-curve of incidence of VAP surpassed the trial sequential monitory boundary. Compared to the standard statistical analysis of metaanalysis, the results of TSA can adjust the false positives or false negatives. No significant association was observed in terms of 90day mortality, overall mortality, 28-day mortality, ICU mortality, hospital mortality, diarrhea, length of ICU stay, length of hospital stay, and duration of mechanical ventilation. VAP is currently the second most common nosocomial infection in America and the most prevalent ICU-acquired infection worldwide. In addition, it is a costly healthcareassociated infection. Rello et al. (2002) suggested that VAP might lead to an additional 40,000 dollar in hospital charges per patient. Branch-Elliman et al. (2015) developed a costbenefit model to determine the most cost-effective strategy for prevention of VAP and examined a total of 120 unique combinations of VAP prevention strategies. They documented that the application of prophylactic probiotics and subglottic endotracheal tubes was cost-effective for prevention of VAP from the perspective of societal and hospital (Branch-Elliman et al., 2015). Combined the results of our present meta-analysis, we concluded that implementation of probiotics for prevention of   VAP in mechanically ventilated patients had the potential to improve the incidence of VAP.
On the topic of VAP prevention in mechanically ventilated patients, four meta-analyses had been performed to evaluate the effectiveness of probiotics (Siempos et al., 2010;Gu et al., 2012;Wang et al., 2013;Bo et al., 2014). Siempos et al. (2010) and Wang et al. (2013) identified five trials, but they yielded an opposite conclusion. Besides, Gu et al. (2012) obtained seven trials and Bo et al. (2014) included eight trials. Compared with the previous meta-analyses, our meta-analysis was largest and most updated, involving 13 trials and 1,969 patients. The results of present meta-analysis were consistent with the two previous meta-analyses (Siempos et al., 2010;Bo et al., 2014), which suggested that probiotics were associated with decreased risk of VAP in mechanically ventilated patients. Furthermore, the present meta-analysis performed a further analysis to confirm   the conclusion. According to the results of TSA, Z-curve of the incidence of VAP surpassed the trial sequential monitoring boundary, indicating that the result of incidence of VAP was true positive. The effect of probiotics in critically ill patients has been evaluated in several studies (Jacobi et al., 2011;Liu et al., 2012;Petrof et al., 2012;Barraud et al., 2013;Manzanares et al., 2016).      They all supported that the use of probiotics could reduce the risk of infection for critically ill patients, including VAP. Therefore, the application of probiotics for VAP prevention should be recommended in clinical practice in the current healthcare circumstance. Several limitations should be taken into consideration when interpreting the results from the present meta-analysis. First, the quality of the included trials relatively low. As shown in Figure 2, even though most of trials adequately reported the methodology, several domains still got "unclear" due to insufficient information in their studies. Second, owing to limited number of included trials, we failed to detect the publication bias, which inevitably affected the precision of our findings. Furthermore, even though we comprehensively searched the databases, the gray literature was not collected. Third, the significant between-study heterogeneity was detected, which might influence the validity of the meta-analysis. The heterogeneity might be derived from the species and dosage of probiotics as well as timing of administration. Ultimately, even though the present meta-analysis is the largest study on this topic, the sample size of the meta-analysis was not large enough. For primary outcome (incidence of VAP), 62.9% of the RIS was accrued and but the cumulative Z-curve has surpassed the trial sequential monitory boundary. For secondary outcomes, however, the cumulative Z-curves neither crossed the futility boundary nor reached RIS. Only 32.0, 15.5, and 25.2% of the RISs were accrued for overall mortality, ICU mortality, and hospital mortality, respectively. Therefore, further trials are needed to verify the conclusion.
In this meta-analysis, we found that probiotics could reduce the incidence of VAP in mechanically ventilated patients. It seems likely that probiotics provide clinical benefits for mechanically ventilated patients. Large sample size and high quality RCTs are needed to further evaluate the effect of probiotics on preventing VAP in mechanically ventilated patients. However, the TSA results of overall mortality, ICU mortality, and hospital mortality showed that there might be false-negative outcomes. Therefore, further trials warranted to identify the value of probiotics in mechanically ventilated patients in future.