INTRODUCTION
Sepsis occurs when the body fails to respond appropriately to infection, leading to fatal organ dysfunction [1]. Although the global incidence of sepsis has steadily decreased since 1990, its high mortality rate and financial burden are still major concerns worldwide [2, 3]. Appropriate initial antibiotic therapy is crucial for reducing sepsis-related mortalities [4, 5], yet not all patients receive adequate antibiotics initially, owing to factors such as clinician perceptions and patient conditions [6–8]. Treatment can be optimized via pathogen identification and antibiotic-susceptibility tests (ASTs). However, conventional methods take 2 or more days to complete [9]. Efforts to reduce this time-to-result (TtR) are ongoing [10] and have achieved promising results [11–14].
The QMAC-dRAST (QuantaMatrix, Seoul, Korea) is a direct AST device of current interest [15–22]. The device performs AST directly on positive blood culture broth (PBCB) within 6 hrs and is based on microfluidic chip technology [23]. Recently, a new version of QMAC-dRAST was released. In contrast with its prototype (QMAC-dRAST V2.0), QMAC-dRAST V2.5 determines minimal inhibitory concentrations (MICs) based on artificial intelligence. Additionally, the preincubation time was reduced from 1 hr to 45 min, and for rapidly growing pathogens, the TtR was decreased from 6 hrs to 4–6 hrs. A built-in expert system is now available for interpreting the determined MICs [19, 20]. In our previous study [24], we demonstrated that AST performed with the bacterial pellet prepared directly from PBCB using SepsiPrep kit gave comparable performance with the conventional AST using bacterial colonies.
To our knowledge, this study is the first to compare both direct AST methods: QMAC-dRAST and BD Phoenix M50 direct AST, i.e., dPhoenix (BD Diagnostic Systems, Sparks, MD, USA), the latter of which employs bacterial pellets prepared directly from PBCB with the SepsiPrep Kit (Asta, Inc., Suwon, Korea). In this retrospective study, we evaluated and compared the accuracy and TtR of both methods. Additionally, we assessed the changes in antibiotics prescribed before and after ASTs were performed.
MATERIALS AND METHODS
Study design
We retrospectively reviewed AST data from patients who visited Seoul St. Mary’s Hospital from January to March of 2024. Only patients for whom AST data from both the QMAC-dRAST and dPhoenix assays were available were included. Patients were excluded if they had a polymicrobial bloodstream infection (BSI) or when bacterial stock was unavailable for further study. This study was approved by the Seoul St. Mary’s Hospital Data Review Board and Institutional Research Board (approval number KC24SISI0698).
Routine laboratory settings for direct identification and direct AST
Our microbiology laboratory operates year-round, and sample preparations for direct identification and direct AST from PBCB are performed three times per day.
Upon arrival, blood culture bottles (BacT/ALERT FA PLUS, BacT/ALERT FN PLUS, and BacT/ALERT PF PLUS; bioMérieux, Marcy L’ Étoile, France) inoculated with patient blood were incubated in the BACT/ALERT VIRTUO (bioMérieux) blood culture system. Aliquots were collected from positive blood culture bottles to perform Gram staining, direct identification, and direct AST.
For direct identification, PBCBs were processed with a SepsiPrep Kit to create bacterial pellets, as previously described [12]. These pellets were then analyzed using one or two matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF MS)-based assays according to the manufacturer’s instructions: ASTA MicroIDSys (Asta, Inc.) and/or Vitek MS (bioMérieux).
Direct ASTs were performed using QMAC-dRAST followed by complementary dPhoenix testing. For QMAC-dRAST, samples were prepared by retrieving 1 mL of PBCB with a sterile syringe and transferring it into a 5 mL polystyrene tube without additional processing. Currently, we use the QMAC-dRAST GN S17 panel in our laboratory, and MICs are interpreted according to the CLSI guidelines and its built-in expert system.
Complementary dPhoenix testing was performed using the aforementioned bacterial pellets instead of isolated colonies grown on solid media. An aliquot of each bacterial pellet was resuspended in BD ID broth (BD Diagnostic Systems) at a concentration of 0.5 McFarland (McF) standards. Next, AST was performed using the NMIC-500 panel, according to the manufacturer’s instructions.
Extended-spectrum-β-lactamase (ESBL) positivity for four species from the order Enterobacterales (Escherichia coli, Klebsiella pneumoniae, Klebsiella oxytoca, and Proteus mirabilis) was evaluated using the QMAC-dRAST system. The ESBL positivity for E. coli, K. pneumoniae, and K. oxytoca was also evaluated using the dPhoenix system.
Broth microdilution (BMD) method
For this study, BMD was performed as a reference method to compare the performance of both direct AST methods, following the 2024 CLSI guidelines [25]. Briefly, pathogens preserved in skim milk were cultured on sheep blood agar (Sinyang Diagnostics, Seoul, Korea), and well-isolated colonies were inoculated in culture medium at a concentration of 0.5 McF standards. Thirteen antibiotics, included in both the QMAC-dRAST and dPhoenix methods, were tested: ampicillin (AM), amikacin (AN), ceftazidime (CAZ), ciprofloxacin (CIP), cefazolin (CZ), ertapenem (ETP), cefepime (FEP), gentamicin (GM), imipenem (IPM), meropenem (MEM), ampicillin–sulbactam (SAM), trimethoprim–sulfamethoxazole (SXT), and piperacillin–tazobactam (TZP). The tests were performed in triplicate and interpreted by two researchers.
Data analysis
The MIC data were interpreted as susceptible (S), intermediate (I) or susceptible-dose dependent (SDD), or resistant (R) according to the 2024 CLSI M100 guidelines. For isolates with intrinsic resistance, the AST results were interpreted as resistant regardless of the MIC value [26].
The AST results obtained using both methods were compared with the BMD results to calculate categorical agreement (CA), very major error (VME; R with BMD testing vs. S with QMAC-dRAST or dPhoenix testing), major error (ME; S with BMD vs. R with QMAC-dRAST or dPhoenix), and minor error (mE; discrepancies occurring when one of the AST methods reported I or SDD). Essential agreement was not calculated because most antibiotics were tested using different concentration ranges depending on the methods used. CA of ESBL positivity was assessed using samples showing comparable data obtained using both the QMAC-dRAST and dPhoenix methods. The acceptable criteria for comparative analysis were as follows: CA ≥90%, VME <3% for all resistant isolates, ME <3% for all susceptible isolates, and mE ≤10% for all the tested isolates [27].
Tendencies for overestimation or underestimation were assessed by comparing the BMD MIC with the MICs of both methods. Differences in the numbers of overestimated cases and underestimated cases were calculated, and the net differences were expressed as ‘↑’ or ‘↓’ arrow symbols for every set of five overestimated or underestimated cases, respectively.
All statistical analyses were performed using Microsoft Excel (Microsoft Corporation, Redmond, WA, USA). P<0.05 was considered statistically significant.
Supplementary tests to resolve discrepancies
For antibiotics with high VMEs determined using both methods, a disk diffusion (DD) test was performed according to CLSI guidelines [28]. Additionally, to exclude the possibility of an inoculum effect or cell damage because of using PBCB or the SepsiPrep Kit, respectively, colony AST was performed for antibiotics showing high VMEs with both methods, as previously described [20].
When isolates showed different ESBL positivity in both methods, inhibitor-potentiation DD testing was performed [26].
In-house <italic>blaOXA−1</italic> PCR analysis
Previous data revealed that automated AST devices showed relatively low CA with regard to TZP, compared with the BMD method [29]. Considering that ceftriaxone-non-susceptible E. coli and K. pneumoniae isolates harboring OXA-1 β-lactamase were a possible cause of this phenomenon [30], we performed an in-house PCR test on isolates for which discordant TZP results were obtained, as described previously [31].
Definition of the TtR
The TtR was defined as the time elapsed from sample arrival at our laboratory to AST completion. Paired Student’s t-tests were performed to evaluate the statistical significance of differences in the TtR values obtained using the QMAC-dRAST and dPhoenix methods.
Changes in antibiotic prescriptions after AST
We reviewed the antibiotics used to treat Enterobacterales-associated BSIs in our study population and classified them into three categories: third- or fourth-generation cephalosporins, β-lactam/β-lactamase-inhibitor combinations, and carbapenems. This classification was based on relevant studies [32, 33] and is detailed in Supplemental Data Fig. S1. Treatment was considered optimal if the pathogen was susceptible to the administered antibiotics according to the AST results. Additionally, optimal treatment specifically entailed administering carbapenems to patients with ESBL-positive isolates and third- or fourth-generation cephalosporins to patients with ESBL-negative isolates.
Only patients with identifiable ESBL data were included. Patients who expired before treatment modification or received medications not listed in Supplemental Data Fig. S1 were excluded.
McNemar’s test was applied to ascertain whether AST facilitated more appropriate antibiotic usage.
RESULTS
Study population and direct identification results
AST data from 118 PBCB inoculates were initially reviewed. Of these, we excluded 11 polymicrobial samples and one sample that was unavailable for BMD. Consequently, 106 specimens were eligible for this study. Direct identification revealed the following distribution: 71 (67.0%) E. coli; 27 (25.5%) K. pneumoniae; three (2.8%) Enterobacter cloacae; and one (0.9%) each of K. oxytoca, P. mirabilis, Raoultella ornithinolytica, Raoultella planticola, and Serratia marcescens.
Performance of both direct AST methods and <italic>blaOXA−1</italic> in-house PCR results
A total of 4,133 MIC results were available for comparative analysis. Both methods (QMAC-dRAST vs. dPhoenix) showed acceptable overall CA (93.6% vs. 93.1%, respectively), VME (1.6% vs. 0.9%, respectively), ME (0.9% vs. 0.6%, respectively), and mE (5.3% vs. 6.2%, respectively) (Table 1). In addition, both methods showed similar trends in overestimating or underestimating MICs for several antibiotics; however, the degree varied depending on the antibiotic (Table 2).
For each method, some antibiotics displayed a CA of less than 90% (CZ, FEP, SAM, and TZP for QMAC-dRAST; CAZ, CIP, CZ, SAM, and TZP for dPhoenix); these results primarily reflected the mEs observed. However, a notably high VME was observed for TZP in both methods (20.0% with QMAC-dRAST and 15.0% with dPhoenix).
ESBL data were available for 100 samples evaluated with QMAC-dRAST; among them, six samples were reported as ‘not determinable.’ Following dPhoenix testing, ESBL data were available for 99 samples, and all samples were classified as either positive or negative. We evaluated the concordance of ESBL positivity using 93 samples, and all samples except one showed identical results (98.9%, 92/93). After DD-based confirmation, the concordance was 100% (93/93) and 98.9% (92/93) for the QMAC-dRAST and dPhoenix data, respectively (Table 1).
Fourteen isolates displayed discrepant TZP results in at least one of the methods tested; QMAC-dRAST showed four VMEs, eight mEs, and two CAs whereas dPhoenix revealed three VMEs, eight mEs, and three CAs. After excluding one sample for which ESBL data were unavailable, we performed DD testing, blaOXA−1 in-house PCR, and AST using colony isolates on the remaining 13 samples (Table 3). Supplementary DD testing confirmed CA in 10 of 13 samples, whereas the remaining three samples showed mEs, with BMD testing showing SDD and DD testing showing R.
Of the 13 isolates, eight harbored both ESBL and OXA-1 β-lactamase, three harbored only ESBL or OXA-1 β-lactamase, and two did not express either of them (Table 3). In-house PCR testing for blaOXA−1 revealed that most E. coli and K. pneumoniae isolates harboring both ESBL and OXA-1 β-lactamase displayed mEs in at least one of the methods tested (87.5%, 7/8).
When compared with the initial TZP MIC results, supplementary ASTs using colonies revealed a few discordant TZP MICs (Table 3). QMAC-dRAST revealed lower TZP MICs in three isolates. With respect to the colony AST of BD Phoenix M50 (cPhoenix), two isolates showed decreased TZP MICs and the other two showed increased TZP MICs.
Clinical impact of direct AST
The QMAC-dRAST results demonstrated an average TtR of 20.8 hrs, which was significantly shorter than the average TtR of the dPhoenix (30.1 hrs; P<0.001).
Among 100 patients with identifiable ESBL data, seven patients passed away before the regimen could be changed, and one patient used antibiotics not listed in Supplemental Data Fig. S1. We reviewed the antibiotic usage and modifications of the remaining 92 patients (Figs. 1 and 2). Before AST, only 12 of 42 patients with ESBL-positive isolates (28.6%) received carbapenems. The remaining 30 patients eventually received carbapenems after AST. Among patients with ESBL-negative isolates, 31 of 50 (62%) initially received third- or fourth- generation cephalosporins. After AST, four of the remaining 19 patients eventually received third- or fourth- generation cephalosporins. However, five of the 31 patients who initially received third- or fourth-generation cephalosporins received antibiotics with a broader spectrum (three received β-lactam/β-lactamase-inhibitor combinations, and two received carbapenems) after AST. Consequently, the number of optimally treated patients increased from 43 (46.7%) to 72 (78.3%) after performing ASTs. By comparing the number of changes from suboptimal treatment to optimal treatment and vice versa, McNemar’s test confirmed that initial suboptimal treatment could be modified to optimal treatment after performing ASTs (P<0.001).
DISCUSSION
QMAC-dRAST testing showed acceptable overall CA, VME, ME, and mE values of 93.6%, 1.6%, 0.9%, and 5.3%, respectively. These results are consistent with those of a previous report showing overall CA, VME, ME, and mE values of 92.1%, 1.8%, 3.5%, and 5.2%, respectively [20]. In another study [23], the VME, ME, and mE values were 1.2%, 2.7%, and 7.9%, respectively, with a lower CA (89.9%).
In terms of individual antibiotics, QMAC-dRAST showed CA values lower than 90% for CZ, FEP, SAM, and TZP. These findings are consistent with previous studies designed to evaluate the QMAC-dRAST with BMD or DD testing, which showed that the CA was lower than 90% for FEP and β-lactam/β-lactamase-inhibitor combinations [16, 19, 23].
To our knowledge, this study is the first to demonstrate the performance of dPhoenix and use of the SepsiPrep Kit to prepare bacterial pellets from PBCB samples. dPhoenix showed good performance, with overall CA, VME, ME, and mE values of 93.1%, 0.9%, 0.6%, and 6.2%, respectively. The results were comparable with those of other studies involving cPhoenix, which showed the following: CA: 96.5%, VME: 0.5%, ME: 0.2%, and mE: 3.3% [34] and CA:97.6%, VME: 1.8%, ME: 0.5%, and mE: 1.7% [35]. However, for individual antibiotics, dPhoenix revealed CA values lower than 90% for CAZ, CIP, CZ, SAM, and TZP. In contrast, other studies designed to compare conventional BMD with cPhoenix did not reveal antibiotics with CA values lower than 90%, despite the fact that the lowest CAs were observed for β-lactam/β-lactamase-inhibitor combinations [34, 35]. Given dPhoenix’s performance and specific error profile, using the SepsiPrep Kit for dPhoenix yielded an acceptable performance. The results are comparable to those of other sample preparation methods for dPhoenix, such as the Sepsityper Kit (Bruker Daltonics, Billerica, MA, USA) [36].
ESBL positivity was comparable in 93 samples; the QMAC-dRAST system correctly determined the ESBL positivity in all 93 samples. These results agree with other findings from studies in which QMAC-dRAST showed no [22] or a few [16] discrepancies, compared with other commercial AST devices. However, in contrast to dPhoenix, QMAC-dRAST did not determine the ESBL positivity in six of 100 samples (6.0%). This discrepancy might have been caused by differences in the antibiotic concentration range of the AST panel or the embedded algorithm for ESBL interpretation.
An alarming finding was that the VME percentage for TZP was high with both methods: 20.0% for QMAC-dRAST and 15.0% for dPhoenix. A high VME (70.6%, 12/17) for TZP was also reported in a study by Wong et al. [16], wherein QMAC-dRAST was compared with DD testing; however, the isolate characteristics were not described. Although high VME values for TZP have been reported for QMAC-dRAST and cPhoenix in some studies [16, 21, 22], the reason for this phenomenon remains unclear. Although mechanisms such as the hyperproduction of classical TEM or SHV β-lactamases [37, 38] and the existence of a combination of classical β-lactamases with porin deficiency [37] have been associated with TZP resistance, we did not elucidate the exact mechanisms underlying TZP resistance. However, we ruled out the possibility of an inoculum effect and cell damage because of the use of PBCB and the SepsiPrep Kit, respectively, using QMAC-dRAST and BD Phoenix M50 to retest all 13 samples showing discrepant TZP results with isolates grown on solid media [20]. We confirmed that OXA-1 β-lactamase was a possible cause of mEs for TZP, considering that most isolates harboring both ESBL and OXA-1 β-lactamase had TZP MICs near the breakpoints established by the CLSI: S and SDD (8–16 μg/mL) for both methods and SDD and R (16–32 μg/mL) for BMD.
In this study, the TtR of our dPhoenix method was expected to be shorter than that of the cPhoenix method as an overnight incubation step was omitted. In laboratories that perform cPhoenix-based assay using isolates obtained after overnight culture on solid media [10, 11, 15], the TtR difference will likely exceed that reported here (9.3 hrs).
With respect to antibiotic changes after performing AST, all patients with ESBL-positive isolates received optimal treatment after AST. Combined with the short TtR, this rapid treatment optimization seems promising. Further, Kim et al. [39] reported that 64.7% of suboptimal treatments were optimized by performing AST and that the use of broad-spectrum antibiotics decreased by 55.9%. However, we found that among patients with ESBL-negative isolates, the number of those that received broad-spectrum antibiotics increased from 19 (38%) to 20 (40%) after AST because of clinical conditions such as sepsis and persistent neutropenic fever (data not shown). Similarly, Anton-Vazquez et al. [33] reported that approximately 75–80% of patients who were eligible for antibiotic modification could not receive antibiotics with a narrower spectrum owing to their clinical conditions.
This study has a few limitations. First, the small number of resistant or susceptible strains for certain antibiotics might have contributed to the high VME and ME rates. Second, the pathogens included were predominantly E. coli and K. pneumoniae. Further studies involving other uncommon pathogens from the order Enterobacterales are required to fully evaluate the accuracy of QMAC-dRAST.
In summary, both QMAC-dRAST and dPhoenix are rapid and relatively accurate methods for performing direct AST. Although direct AST has several limitations, such as the requirement for rapid identification of bacterial isolates (e.g., MALDI-TOF MS or nucleic acid amplification) and its inapplicability to polymicrobial samples, it is still expected to improve the clinical outcomes of patients with Enterobacterales-associated BSIs when appropriately implemented.