Bloodstream infections (BSI) are a serious clinical adverse events associated with very high morbidity and mortality. They represent one of the most common health care related infections and rank in the top four most costly conditions. The diagnosis of BSI is based upon blood cultures, with automated detection of the presence of viable microorganisms. Matrix-assisted laser desorption/ionization time-of-flight (MALDI-TOF) mass spectrometry can allow quick identification of the responsible organism after its culture (within 24 h after the positive flag) (Bizzini et al., 2010; Greub, 2010) but it could also be performed directly from positive blood cultures (Prod'hom et al., 2010; Clerc et al., 2013), thus saving considerable time. Recently, Marschal et al. have described the Accelerate Pheno system as an automated test system capable of performing both identification and AST directly from positive blood cultures of Gram-negative organisms within approximately 7 h (Marschal et al., 2017). Near-future alternatives of antimicrobial susceptibility tests with answer less than 5 h include MALDI-TOF, due to its ability to detect some mechanisms of resistance (Burckhardt and Zimmermann, 2011), and flow cytometry (Alvarez-Barrientos et al., 2000; Pina-Vaz and Rodrigues, 2010). Molecular methods can also contribute to a rapid and specific identification of microorganisms. However, concerning susceptibility testing, scarce useful information can be provided since only a few genes codifying for resistance-associated proteins are yet known (Salimnia et al., 2014; Ceyssens et al., 2016). A global susceptibility profile is impossible to be obtained by molecular biology unless whole genome sequencing tools are used (Zankari et al., 2013).

The emergence and spread of antimicrobial resistance constitutes a major medical problem. Laboratory assessment of antimicrobial susceptibility is a prerequisite for adequate management of infections. However, prolonged time is required for routine phenotypic antimicrobial susceptibility testing (AST) since growth in the presence of different antimicrobial drugs is mandatory. Flow cytometry (FC) represents an efficient and fast approach for the analysis of cell architecture and functional phenotypes, with considerable advantages over conventional methods. By studying structural lesions induced by drugs that lead to changes in morpho-functional parameters (e.g., membrane potential, cell size, amount of DNA), antimicrobial susceptibility can be assessed without the need of prolonged microbial growth. The availability of such an assay for susceptibility testing in critical bloodstream infections is hereby proposed, since it can be performed in just a few hours, with high accuracy. We previously described microbiological applications of FC, including susceptibility evaluation of antifungals, as well as the elucidation of antifungal resistance mechanisms (Pina-Vaz and Rodrigues, 2010). Moreover, we developed AST cytometric protocols regarding the main mechanisms of resistance to betalactams with the most important bacteria involved in clinical infections (Faria-Ramos et al., 2013; Silva et al., 2016). In this study we evaluate the performance of FASTinov^® kit for AST of Gram negative bacteria from positive blood cultures. In collaboration with members of the antimicrobial stewardship program of our university hospital, we also determined if the anticipation of AST results (less 46 h) would have led to changes in empiric antibiotic therapy and, consequently, de-escalation of treatment.

Table 1 aggregates data from patients included in the study regarding, age, gender, hospital department of admission, length of staying (LOS), empiric antimicrobial therapy and clinical outcome. Almost 17% of the patients received inappropriate therapy, which was associated with a 16-day increase of hospital LOS. In 10 patients receiving initially inadequate therapy and in whom antibiotics were changed, such change occurred, on average 2.8 days after the initial prescription and was made either before the routine AST result was available (suggesting that this decision was based on risk factors and clinical evolution) or (5/10 cases) on the same day that AST results became available. In this last instance, patients received inappropriate therapy during 4.2 days, on average. De-escalation was deemed possible in 65 out of 85 (76.5%) patients, but was done in only 23 (27.1%) of them. De-escalation occurred, on average, 4.5 days (0–10) after the start of antibiotic therapy and in most cases (20/23) only after AST result was known. Regarding the outcome, data analysis revealed no significant differences among patients who had received appropriate/ inappropriate therapy (p = 0.455).

The drug most frequently used was piperacillin-tazobactam (27.4%) followed by ceftriaxone (24.5%); when combination therapy was used, piperacillin-tazobactam plus amikacin (4.9%), imipenem plus vancomycin (4.9%) and piperacillin-tazobactam associated with vancomycin and amikacin (2%) were the most frequent drug combinations.

Regarding AST time-to-result, the routine laboratory method (Vitek2) required on average 51 h (48–58 h). This range of time was often associated with the need to perform additional susceptibility testing, such as agar disk diffusion or Etest in several clinical isolates to confirm results. FASTinov^® results were available within a range of 2 h.

Table 2 details the Gram-negative bacilli studied (isolated from blood cultures and those spiked with selected control strains), the respective resistance phenotype, the number of ESBL, carbapenemases and AmpC positive strains, and the number of multiresistant strains based on ISO microdilution methods. Regarding the clinical isolates, 88.2% were Enterobacteriaceae (60% E. coli, 18.9% K. pneumoniae, 3% Enterobacter spp., 1% Citrobacter spp 2% Proteus spp. and 3% Serratia spp). Regarding non-Enterobacteriaceae, Pseudomonas spp were the most isolated species followed by Acinetobacter spp (9.8 and 1%, respectively).

The FASTinov^® gramneg kit results are detailed in Table 3. The highest CA was observed for colistin (1.00), followed by piperacillin/tazobactam, gentamicin, meropenem and cefotaxime (0.98). A low CA was detected for ciprofloxacin (0.94), with a major error rate of 3.57%. A high major error rate was also observed for amoxicillin-clavulanic acid (3.80%). The very major discrepancies were verified for meropenem (6.25%), gentamicin (2.78%), and amoxicillin-clavulanic acid (2.00%). Regarding ESBL detection, the CA between FASTinov^® gramneg kit and ISO microdilution was 1.00.

We found a higher agreement between the gold standard and FASTinov^® kit (0.98) than between the gold standard and Vitek2 (0.97).

Our study clearly evaluates the performance of FASTinov^® gramneg kit for AST directly from positive blood cultures of Gram-negative bacteria. The overall categorical agreement between FAStinov^® test and the ISO reference method was 98%, with a rate of minor, major and very major discrepancies of 1.03, 1.59, and 1.06%, respectively. According with Food and Drug Administration (FDA), the rate of major discrepancies attributable to a test device should be < 1.5% for individual drug comparison; the rate of major discrepancies should not exceed 3%, and the overall categorical agreement should be >90% (CDRH-Guidance, 2009). The very major discrepancies obtained with FASTinov^® gramneg kit were detected for amoxicillin-clavulanic acid (2.00%), gentamicin (2.78%) and meropenem (6.25%). In case of gentamicin and meropenem, the high VME rates are related with the low number of resistant strains tested (1/36 and 1/16, respectively), and represents a limitation of this study. Notably, the FASTinov^® gramneg kit correctly detect resistance phenotypes to amikacin, cefotaxime, ceftazidime, ciprofloxacin, colistin, and piperacillin-tazobactam, not providing false susceptible results. Regarding major discrepancies, high rates occurred for amoxicillin-clavulanic acid (3.80%) and ciprofloxacin (3.57%), exceeding the limits proposed by the FDA. Curiously, Vitek2 system also provided false resistant results for CIP, with a rate of major discrepancies of 3.57% (data not shown). In case of piperacillin-tazobactam, only one major (1.15%) and one minor (0.78%) discrepancies were observed with FASTinov^® gramneg kit. Some authors have reported serious rates of errors for TZP with automated systems, namely BD Phoenix, MicroScan WalkAway, Vitek, and Vitek2 (Sader et al., 2006; Juretschko et al., 2007). Simultaneously, the new Accelerate Pheno System also display high rates of VM errors for TZP (8.2%).

FASTinov^® gramneg kit also provided a correct detection of ESBL-producing strains (CA = 1.00). Thus, these results demonstrated the potential of FASTinov^® gramneg kit for detection of resistance phenotypes with a timely result within 2 h.

Due to the impossibility of a faster AST result, clinicians are frequently confronted with the need to prescribe wide spectrum antibiotics in order to minimize the risk of inappropriate therapy. However, this strategy significantly increases the selective pressure upon microorganisms and might ultimately lead to the emergence of antimicrobial resistance (Paiva, 2013). Additionally, in case of severe infections, e.g., sepsis, the time interval until the beginning of appropriate antibiotic therapy has been shown to correlate closely with outcome (Kumar et al., 2009). Kollef and Micek have suggested that, in critically ill septic patients, broad spectrum therapy should be only prescribed when considered clinically necessary, while maximizing the possibility of obtaining a valid microbiological diagnosis—de-escalation (Kollef and Micek, 2005). As soon as clinical improvement is noted and/or a culprit microorganism is identified, clinicians should then narrow the spectrum of such empiric therapy to cover only the target microorganism. However, it has been suggested that for these strategies to be effective they should be implemented as early as possible (Weiss et al., 2015). Thus, the development of laboratorial tests able to provide early information about antimicrobial susceptibility profile is urgently needed. These tests would carry the dual benefit of limiting exposure to initially inappropriate therapy and the added risks involved, while allowing for faster hence more effective de-escalation. Recently, several methodologies like PCR-based techniques, mass spectrometry, microarrays, microfluidics, cell lysis-based assays and whole-genome sequencing have all demonstrated similar ability to detect resistance in various bacterial species but with several limitations, with phenotypic assays showing superior accuracy (Pulido et al., 2013).

The FASTinov^® gramneg kit had the advantage of providing phenotypic and much earlier information about AST. The time-to-result of the FASTinov^® method was significantly shorter (2 h vs. a minimum of 24 h), potentially allowing an earlier switch to a narrow spectrum, targeted therapy, thus reducing selection pressure and minimizing exposure to inappropriate therapy and its side effects. As the most microorganisms isolated from blood cultures in our study were susceptible to the antimicrobials tested, de-escalation would be possible in about 80% of the cases. However, in the scenario of routine AST (around 48 h after the positive flag of the blood culture), with Vitek2 de-escalation was only performed in 27% of the cases.

Although the EUCAST breakpoints are species dependent, AST with FASTinov^® gramneg was performed without previous identification (only based on Gram stain), except in case of control strains. It was possible because the clinical isolates included in this study were very susceptible (most of strains) or very resistant to the tested drugs. However, in case of strains with MIC values closer to breakpoint concentrations FASTinov software will not inform about the phenotype before strain identification.

In summary, we found that the FASTinov^® kit shows excellent categorical agreement and low number of errors when compared with both the gold-standard broth-microdilution protocol and the widely used Vitek2 system. The ability of FASTinov^® to promote optimal use of antibiotics deserves further testing. Added benefits might arise from reduced hospital LOS, lower consumption of antibiotics and improved infection control. Large studies are needed in order to reveal the real impact of new diagnostic tools in antibiotic stewardship. Nevertheless, the FASTinov^® gramneg method seems to be a valuable tool to perform the AST in clinical routine.

SC-d-O, CP-V, and AR contributed with the experimental design of this study. AS-D, RT-S, NM, IM-O, and AS carried out all experiments. RT-S, SC-d-O, and CP-V analyzed and interpreted the results. EP, PM, JP, and RC evaluated and discussed the potential impact of the results. SC-d-O, RT-S, and CP-V wrote the manuscript, and all authors performed a critical revision and approved the final version.