This systematic review provides a comprehensive and critical evaluation of the current evidence on the available antibiotic options for treating patients with sepsis due to CR A. baumannii, Enterobacterales, and P. aeruginosa. Our results clearly revealed the challenges to developing evidence-based recommendations on the therapeutic management of severe infections caused by CR-GNB from the existing literature. The included studies showed a suboptimal reporting of key-variables, making results difficult to compare and ill-suited for adjusted analysis. Despite the availability of numerical data on 11546 patients, no quantitative analyses could be performed because of the vast heterogeneity and the lack of clear definitions of the included antibiotic regimens.
The first part of our systematic review focused on a descriptive analysis of clinical and microbiological data. In patients with CR-GNB infections, especially those with critical illness, comorbidities and baseline severity of disease are known to be major contributors to the final outcomes.13 However, our systematic review showed that only 21 comparative observational studies included an adjusted analysis for these confounders, whereas the remaining studies were generally too small to allow for adjustment. Overall, the studies had a median sample of 49, a figure which is considerably lower than the estimated sample of several hundreds or even thousands of patients that are needed for reliably assessing independent association of one antibiotic regimen with mortality in observational studies.6,14
In terms of microbiological characteristics, causative pathogens varied in terms of species, mechanisms of resistance, minimum inhibitory concentrations thresholds to carbapenems, and pattern of concomitant resistance. This heterogeneity was common not only among different studies, but often within the same study. Overall, resistance to carbapenem was detected using five microbiological methods with various performance and accuracy.15
Evidence from observational studies shows a survival benefit in patients with severe infections receiving in vitro active empirical treatment.16,17 However, a post hoc analysis of the only available RCT on the topic documented that early initiation of colistin did not impact on survival of patients with severe infections sustained by colistin-susceptible CR-GNB.18 Similarly to other clinically relevant variables, data on empirical coverage from our systematic review were scarce (only 23% of the studies) and did not allow to conduct any further analysis.
Our systematic review included mostly low-quality observational studies. The included randomized controlled trials were generally of better quality, but did not contribute significantly to the overall analysis. In particular, the trials on new antibiotics showed important limitations related to the inclusion of very small sample size and the use of heterogeneous comparison groups.19,20
In the last years, several systematic reviews and meta-analyses have addressed the issue of the superiority of combination treatment, mainly focusing on polymyxin-carbapenem combination. Conclusions from these systematic reviews were generally undermined by methodological limitations of the included studies 6,21,22 and they did not match the results from a more recent RCT addressing the same clinical question.23 Our systematic review included overall 97 distinct antibiotic regimens; this result reflects the lack of standardization in clinicians’ prescribing, as pointed out in the third paper of this series (reference to be added). Among these 92 regimens, less than a half were accurately defined. Most studies, in fact, adopted a generic definition of combination therapy frequently with unknown in vitro susceptibility, rather than defining specific combinations. In this way, patients could be easily allocated to the various treatment arms and potentially contribute to the retrospective assessment of determinants of mortality. However, in the wake of existing evidence and experts’ recommendations, we deemed those data as scarcely informative for clinical practice and we restricted the analysis to clearly defined antibiotic regimens.21
The purpose of our quantitative analysis was to evaluate the effectiveness of defined antibiotic regimens in patients with sepsis due to CR A. baumannii, Enterobacterales and P. aeruginosa in terms of clinical and microbiological outcomes. Given that several studies did not offer a direct comparison, we selected the NMA approach that can integrate both direct evidence (from studies directly comparing two antibiotic regimens) with indirect evidence (information on two antibiotic regimens derived via a common comparator), increase the precision in the estimates and produce a relative ranking of all antibiotic regimens for selected outcomes.24 Unfortunately, the distribution of the patients across 45 defined antibiotic regimens did not allow the creation of a ‘connected network’ where all regimens are comparable via direct or indirect comparison. We could have forced the analysis or conducted separate NMAs for each network component, but potential results would have been again of poor clinical significance.