Background
Nosocomial pneumonia is a leading cause of death in critically ill patients [
1,
2]. The mortality rates associated with hospital-acquired pneumonia (HAP) or ventilator-associated pneumonia (VAP) in intensive care units (ICU) range from 38% to 70% or higher [
3,
4]. One of the most common pathogens of nosocomial pneumonia is
Acinetobacter baumannii (
A. baumannii). Because
A. baumannii exhibits relatively high virulence and antimicrobial resistance compared with other organisms, the prevalence of multidrug-resistant (MDR) or extensively drug-resistant (XDR)
A. baumannii has kept increasing to at least 80% in past decades [
5]. One of the important risk factors for MDR/XDR-bacterial infections in ICU is an inappropriate antibiotic therapy [
6]. Considering the paucity of clinical data on the comparative effectiveness of various antimicrobial treatments for MDR/XDR
A. baumannii pneumonia, it is pressing to accumulate clinically reliable scientific evidence to guide the selection of an optimal antimicrobial treatment for the infection.
Colistin-based, sulbactam-based and tigecycline-based antimicrobial treatments are currently considered the reserved treatment options for MDR/XDR
A. baumannii infections [
7,
8]. However, the clinical data comparing the antimicrobial treatments adequately with respect to the effectiveness and safety of the treatment of such infections, particularly nosocomial pneumonia, are inconsistent owing to small sample sizes, and substantial between-study heterogeneity [
9‐
13]. In addition, to date there is no consensus based on strong evidence to confirm the therapeutic superiority of a monotherapy or combination therapy and clinical preferences among various antimicrobial combination regimens [
9,
10]. Therefore, clinicians are still facing challenges to apply evidence-based pharmacotherapy in clinical practice for the treatment of nosocomial pneumonia.
Currently available studies have compared the effectiveness of different antimicrobial treatment options only in pairs within the studies [
14‐
16]. Although the results from those studies are informative, the relative effectiveness throughout a variety of therapeutic options remains unknown. Network meta-analysis (NMA) is a relatively novel meta-analysis strategy that integrates both direct evidence from pairwise comparisons within a study and indirect evidence from common-comparator comparisons across the studies [
17,
18]. Compared with conventional meta-analyses, NMA allows comparisons across multiple treatments simultaneously even when the treatments were not directly compared in previous studies. Furthermore, NMA using a Bayesian approach uniquely provides the probability estimates that enable clinicians to make an intuitive pharmacotherapy decision [
19]. The Bayesian approach allows us to adopt the strengths from data across multiple studies and does not require an assumption of normal distribution [
20]. Therefore, the approach is more favorable than the frequentist approach when small numbers of studies are included in each pair of comparisons.
The aim of this study was to evaluate the comparative effectiveness of currently available antimicrobial options, including monotherapy and combination therapy, for the treatment of critically ill patients with nosocomial pneumonia caused by MDR/XDR A. baumannii, using a Bayesian NMA approach.
Methods
A systematic review and the Bayesian NMA were performed in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) extension statement for network meta-analyses [
21].
Search strategy
A comprehensive literature search was performed using the electronic databases of PubMed, Embase, and the Cochrane Central Register of Controlled Trials from the inception of each database to 31 March 2017. The literature search was limited to human studies without language restrictions. In order to identify appropriate articles that evaluated the antimicrobial treatments for patients with nosocomial pneumonia caused by MDR/XDR pathogens, the following text words and medical subject heading (MeSH) terms were used: “pneumonia”, “HAP”, or “VAP”; “drug resistant”, “carbapenem-resistant”, “MDR”, or “XDR”; “aminoglycoside”, “carbapenems”, “colistin”, “fosfomycin”, “glycopeptide”, “glycylcycline”, “polymyxin”, “rifampin”, “sulbactam”, or “tigecycline”. In addition, proceedings from relevant conferences and the references that were listed in all retrieved articles were also manually searched to ensure a complete identification of all eligible studies.
Selection criteria
The types of study included in the analysis were randomized controlled trials (RCTs) and observational studies in which antimicrobial agents were compared in the treatment of critically ill adult patients with HAP or VAP caused by MDR/XDR A. baumannii. Studies were included if they evaluated at least one of the following three outcomes with clear definitions: all-cause mortality, clinical cure, or microbiological eradication. All-cause mortality, defined as the incidence of deaths from any cause in ICU within the follow-up duration of approximately 30 days, was chosen as the primary outcome variable for the comparison of antimicrobial effectiveness. If all-cause mortality in the ICU was not available in the studies, in-hospital mortality data were used. Clinical cure and microbiological eradication were selected as the secondary outcome variables. Clinical cure was defined as a resolution of signs and symptoms of pneumonia by the end of therapy, whereas microbiological eradication was defined as a confirmed negative result in a follow-up bacterial culture by the end of therapy.
Excluded studies are as follows: (1) case series, (2) studies that enrolled pediatric patients, (3) studies that enrolled patients with community-acquired pneumonia, and (4) studies in which less than half of study population had pneumonia.
Quality assessment and data extraction
Two investigators (SYJ and SHL) independently screened the titles and abstracts of the articles, and reviewed their full texts based on the pre-specified selection criteria. Any inconsistencies between the two investigators were resolved by extensive discussion with the third investigator (HN). Quality assessment of included studies was performed by two investigators (SYJ and SHL), and was crosschecked by the third investigator (SY). The quality of included RCTs was assessed using the Cochrane Collaboration Risk of Bias Tool [
22]. The quality of observational studies was evaluated using the Newcastle-Ottawa Scale (NOS) [
23]. Studies with an NOS score <7 were considered at high risk of bias [
24,
25]. After excluding the studies with high risk of bias, the remaining studies were included in the final analyses. Two investigators (SYJ and SHL) independently extracted the following data from each study using a standardized extraction form: year of publication, type of study design (i.e., RCT or observational study), patient population characteristics, specific antimicrobial therapy with dosing strategy as either an intervention or a comparator, and outcome measurements with their definitions. Antimicrobials administrated at greater than standard dosage for the treatment of MDR/XDR
A. baumannii pneumonia were separately categorized as high-dose regimens for each antimicrobial agent, as referred to in previous studies [
26‐
28].
Data synthesis and analysis
Pairwise meta-analyses were initially performed for direct comparisons between different antimicrobial treatments using Stata software (version 13.0, StataCorp, College Station, TX, USA). Homogeneity and consistency were assumed in drawing valid conclusions from NMA analyses [
29]. Homogeneity assumption was satisfied when the magnitude of heterogeneity within direct pairwise comparisons was acceptable. Heterogeneity of the treatment effects across trials in each pair was examined by the
Q test and quantified using the
I
2 statistic. The
I
2 values < 40%, 40%–75%, and > 75% accompanied by a
p value <0.10 from the
Q test were considered mild, moderate, and high heterogeneity, respectively [
30]. A consistency assumption referring to the lack of disagreements between direct and indirect comparisons was required to be met in integrating direct and indirect evidence in the NMA [
31]. Inconsistency in the entire network of each outcome was assessed by either the Lu-Ades model or design-by-treatment interaction model, according to a configuration of the loop in the network [
32‐
34]. The consistency assumption was rejected when the
p value of the inconsistency test was < 0.05. Publication bias was also evaluated with a funnel plot and Egger’s test if ten or more studies were included in each pairwise meta-analysis [
30].
The Bayesian NMA was performed for multiple comparisons using WinBUGS (version 1.4.3, MRC Biostatistics Unit, Cambridge, UK). Both fixed-effect and random-effect models were fitted. The best model was chosen based on the deviance information criterion (DIC) that suggests a significantly better fit of the model with a value lowered by 2–3 points [
35]. Markov chain Monte Carlo (MCMC) samplers were run in WinBUGS using three chains with different initial values. Non-informative priors were used to produce posterior distribution for the treatment effects, allowing the data to dominate the final estimates (vague normal distribution with mean of 0 and variance of 0.0001). There were 10,000 updates generated for each set of the chains, and the first 10,000 iterations were discarded as the burn-in phase. Brooks-Gelman-Rubin diagnostic plots were used to verify the convergence of the MCMC simulations [
36].
The estimates of Bayesian NMA were reported as rank probabilities to identify the relative rankings of antimicrobial treatments based on the surface under the cumulative ranking curve (SUCRA), ranging from 0% (statistically certain to be the worst treatment) to 100% (statistically certain to be the best treatment) [
37]. Because SUCRA rankings could exaggerate the small differences since those are relative values, the estimates of median outcome rates were also reported with the corresponding 95% credible intervals (CrIs) to specify the absolute magnitude of therapeutic effectiveness. Then, Bayesian posterior probabilities of superiority were calculated to identify a significant difference between an individual treatment and a common comparator. The probabilities of superiority (
P) indicated the probabilities of the odds ratio (OR) being < 1 for all-cause mortality and > 1 for clinical cure and microbiological eradication. The treatment with
P > 97.5% or
P < 2.5% was considered statistically superior or inferior to the comparator, respectively [
38‐
40]. Although whether or not a confidence interval (CI) for the OR not crosses 1 determines a statistically significant difference in frequentist statistics, the concept of posterior probability in Bayesian statistics was used to demonstrate the certainty of comparative results [
41‐
43].
Sensitivity analyses were performed to evaluate the robustness of the Bayesian NMA results. Although most patients included in this analysis had drug-resistant A. baumannii pneumonia, the comparative effectiveness of antimicrobial treatments could be affected by type(s) of infection other than pneumonia or causative pathogen(s) other than A. baumannii. Therefore, two sensitivity analyses were performed exclusively using the studies in which all patients had pneumonia or were infected by A. baumannii, respectively.
Discussion
To the best of our knowledge, this is the first and the most comprehensive Bayesian NMA evaluating the comparative effectiveness of various antimicrobial treatment regimens for MDR/XDR
A. baumannii pneumonia in critically ill patients. An important finding of our study is that SUL was the most effective therapy to reduce the all-cause mortality in critically ill patients. The top five treatments with high probabilities of survival benefit were SUL (SUCRA 100.0%), HD SUL (85.7%), FOS + IV COL (78.6%), IH COL + IV COL (71.4%), and HD TIG (71.4%) (Fig.
3). Those five treatment options generally ranked high for improving clinical cure and microbiological eradication as well. However, HD TIG and HD SUL had relatively lower SUCRA values for microbiological eradication among the 15 different antimicrobial treatments. A possible explanation for those lower rankings in microbiological eradication is that, based on the published data,
A. baumannii isolates in the two HD treatment groups were less susceptible to TIG or SUL (MIC >1 mg/L for TIG; MIC >16 mg/L for SUL, respectively) [
49,
52].
The results of this study corroborate growing recent evidence that suggests SUL as a promising treatment option in the management of
Acinetobacter infections [
67‐
69]. Multiple clinical studies have reported that the patient group treated with SUL had a substantially low rate of mortality, ranging from 17% to 33% during approximately 2 weeks of treatment [
67‐
70]. In addition, Oliveira et al. report that polymyxin (colistin or polymyxin B) treatment is significantly associated with higher mortality than SUL, with a relative risk of 1.52 [
70]. Similarly, this Bayesian NMA demonstrated that SUL was superior to IV COL with the probabilities of superiority greater than 97.5% in terms of reducing all-cause mortality and improving microbiological eradication. However, caution needs to be exercised when interpreting and applying our findings to clinical practice, owing to the retrospective nature and relatively small sample sizes of the studies included in this NMA, and potential inherent bias, if any, in reporting the results of the original studies.
According to a recent conference report from the European Society of Intensive Care Medicine, a 4-hour infusion of SUL 3–4 g every 8 hours is recommended for severe
A. baumannii infections involving isolates with higher MICs for SUL (≥8 mg/L) [
10]. A recent pharmacodynamic modeling study conducted in healthy adults demonstrated that a 4-hour infusion of SUL 3 g every 8 hours would be an appropriate dosage regimen of SUL for less-susceptible
A. baumannii [
71]. The findings from healthy adults may not be generalizable to critically ill patients owing to pharmacokinetic alterations associated with critical illness [
71,
72]. Nevertheless, it is noteworthy that prolonged-infusion dosing was found to be a much more effective strategy to achieve a high probability of target concentration attainment over a range of MICs than a dose-escalation strategy [
71]. In the HD SUL treatment group in this NMA, the infusion time of HD SUL was within 1 hour even though SUL MIC for isolated
A. baumannii was > 16 mg/L [
49]. It could be inferred that a more prolonged infusion time was necessary for improving microbiological eradication, considering the time-dependent antimicrobial activity of SUL [
72].
Among combination regimens evaluated in this NMA, FOS + IV COL and IH COL + IV COL had a more beneficial effect on all-cause mortality, with favorable effectiveness in clinical cure and microbiological eradication (Fig.
6). In terms of microbiological eradication, FOS + IV COL demonstrated the greatest SUCRA value in our Bayesian NMA. FOS may be an effective adjunctive therapy for pneumonia caused by MDR/XDR
A. baumannii, considering the synergistic effect of COL and FOS in vitro [
73‐
75]. However, owing to the paucity of clinical data evaluating the efficacy and safety of FOS + IV COL, adjunctive FOS should be used with caution until clinical studies adequately confirm its promising effect in patients with severe pneumonia. According to the guideline recently updated by the Infectious Diseases Society of America, adjunct IH COL therapy is suggested for the treatment of HAP/VAP due to
A. baumannii that is sensitive only to COL, particularly for patients with insufficient response to IV COL monotherapy [
76]. Similarly, in this NMA, the probability of superiority analysis showed that IH COL + IV COL yielded additional therapeutic superiority in terms of clinical cure, and comparable effectiveness in other outcomes, compared with IV COL.
Besides IH COL, several aerosolized antimicrobial agents including amikacin, tobramycin, and fosfomycin were evaluated for the treatment of gram-negative pneumonia in previous studies comparing the efficacy of IH antimicrobial adjunct therapy to various IV antimicrobial agents with that of IV monotherapy or IH placebo [
77‐
80]. However, antimicrobial agents evaluated in those studies could not be included as treatment nodes in our NMA to construct a single connected network. Furthermore, non-MDR/XDR
A. baumannii pneumonia patients were largely included in those studies. Of note, more evidence is required to determine the appropriate administration method of IH antimicrobials for optimal clinical benefit in patients with pneumonia because the delivery devices, dosing, ventilator settings, and endotracheal tube size may affect the therapeutic response [
76,
81].
There are a few limitations in this meta-analysis. The major limitation of this NMA is the absence of aerosolized antimicrobials other than IH COL as aforementioned. Another important limitation is that the analysis was mostly based on retrospective studies (16 out of 23 studies in total). Considering the retrospective nature of most included studies, this NMA should be viewed as a hypothesis-generating study rather than definitive clinical evidence, despite rigorous quality assessments. The safety profiles of all antimicrobial treatments were not evaluated due to insufficient data on adverse drug reactions and substantial differences between studies in the baseline laboratory values or organ function parameters. In addition, the novel antimicrobial treatments with potential activity against A.
baumannii, such as vabomere, plazomicin, cefiderocol and eravacycline, were not included in this NMA owing to absence of published data on the patients with MDR/XDR
A. baumannii pneumonia [
82‐
84]. It remains to be determined whether any of these agents may have a role in this clinically important infection. Additionally, this NMA could not clearly address the role of combination therapy over single-agent therapy. In fact, the purpose of this NMA was not to specifically compare the effectiveness between combination therapy and monotherapy but to evaluate the overall comparative effectiveness of different antimicrobial treatment options. Last, variability in identifying the causative pathogen among the included studies could not completely distinguish colonization from infection in HAP and VAP. Clinical studies more robustly detecting the true infectious organism in HAP or VAP may be necessary to accurately and precisely determine the effects of antimicrobial treatments for MDR/XDR
A. baumannii pneumonia.