Potential influence of rapid diagnostics on timeliness of management decisions for patients with positive blood cultures

Abstract

Background

The Accelerate Pheno system (AXDX) provides rapid identification (ID; 90 minutes) and antimicrobial susceptibility testing (AST; approximately 7 hours) from positive blood culture (BC) bottles. We assessed the potential of AXDX results to influence more timely antibiotic interventions with a convenience sample of 158 positive BCs.

Methods

BCs with a mono-microbial Gram stain likely to be on the AXDX panel were run in parallel with the standard of care (SOC). Using results from the SOC, the medical microbiologist on call (MMOC) noted interventions made at the time of BC Gram stain and when ID and AST results were available. The timing of MMOC intervention was noted and compared with fastest potential SOC time and AXDX time.

Results

Of 158 specimens selected for analysis, 144 were evaluable. ID was available 11.9 hours and AST 27.7 hours faster than SOC. Correct ID was provided for 85.2% of specimens and AST for 59.0% of specimens, with 97.5% essential agreement compared with the SOC. One hundred and thirteen clinical interventions were made on 100 specimens: 54.9% were narrowing; 33.6%, escalation; 6.2%, consultation with ID; and 3.5%, further investigation. If AXDX data had been used immediately once available, interventions would have been possible 24 hours earlier for ID interventions and 39 hours earlier for AST results.

Conclusions

Results from rapid diagnostic panels such as AXDX have the potential to support timely antimicrobial de-escalation and other decisions to benefit patients, especially if paired with stewardship interventions.

Sepsis and bacteremia are common causes of in-hospital death worldwide, causing 25% and 22% mortality, respectively (1,2). Positive outcomes with sepsis and bacteremia are correlated with rapid therapy (3,4), which guidelines recommend be broad spectrum while awaiting results of diagnostic testing (2,5). Broad empiric coverage increases the incidence of antimicrobial resistance, and judicious use of antimicrobials reduces superinfections, shortens length of stay, and reduces readmissions (6–10). Thus, the ideal management of bacteremia would cover the pathogen, without collateral damage to the patient’s healthy flora. Technology can reduce the time to receive antimicrobial susceptibility testing (AST) results, which could hasten the use of appropriate antibiotics. The Accelerate Pheno system (AXDX; Accelerate Diagnostics, Inc., Tucson, Arizona) reduces the time to identification (ID) of common organisms found in blood cultures (BCs) to 90 minutes and reduces AST to approximately 7 hours, compared with 18–24 hours for ID and up to 72 hours for AST with routine laboratory methods. The ability of this improved turnaround time to affect changes in prescription is unknown and likely varies depending on local susceptibility patterns, infrastructure, and prescribing habits. The objective of this study was to determine whether use of AXDX could affect patient management, specifically antimicrobial prescribing, compared with the standard of care (SOC) in a Canadian teaching hospital.

Methods

Setting

This study was performed at a microbiology lab that provides BCs for five acute care hospitals in Vancouver, British Columbia, covering approximately 1,300 beds in total.

Microbiologic testing

BCs were performed using BACTEC Lytic/10 Anaerobic/F and plus aerobic/F medium and incubated in the BACTEC FX automatic BC system (Becton Dickinson, Franklin, New Jersey). A routine BC set at our centre included two separate punctures, with an aerobic and anaerobic bottle drawn for each (four bottles total). A convenience sample of positive BCs was selected, based on technologist and machine availability. Specimens were included if they had the following Gram stain findings: mono-microbial cultures with enteric Gram-negative rods, Gram-positive cocci, or oval yeast. Only one specimen per patient was tested in new bacteraemic episodes (i.e., the first culture set to be positive of the observed morphology). Excluded samples were those that were either poly-microbial on Gram stain or had a morphology other than Gram-negative rods, Gram-positive cocci, and oval yeast. Specimens were also excluded if data were not recorded or if the AXDX testing cartridge had been stored out of temperature range. Samples were run on AXDX and SOC in parallel, immediately after Gram stain reading.

The SOC consisted of subculture onto solid media and rapid matrix-assisted laser desorption/ionization time-of-flight (MALDI-ToF) testing (Bruker, Biotyper Microflex LT/SH library nos. 5627 and 6903; Becton Dickinson, Franklin, New Jersey) at 4–8 hours of incubation, if visible growth was present on routine plate review done at regular hours (7:00 a.m., 1:00 p.m., 5:30 p.m., and 12:00 a.m.) and an acceptable ID was obtained, or standard MALDI-ToF at 18–24 hours if the early MALDI-ToF did not produce a result. The earliest valid result (from either early or routine MALDI-ToF) was used as the SOC time to ID.

Susceptibility testing was performed using the Phoenix PMIC-84 (Gram-positive) and NMIC-404 (Gram-negative) panels run on a Phoenix AP 100 (Becton Dickinson, Franklin, New Jersey). Clinical and Laboratory Standards Institute (CLSI) interpretive breakpoints were used, based on the 2017 M100 document (11). Organism ID was primarily performed with MALDI-ToF and occasionally with API strips (Biomérieux, Marcy-l’Étoile, France). When standard methods failed to provide reliable ID, 16S polymerase chain reaction was performed by a reference lab (British Columbia Centre for Disease Control Public Health Laboratory, Vancouver, British Columbia).

Selected blood samples were tested following manufacturer’s instructions for AXDX and using software versions 1.2.1.22 and 1.3.0.22. Results of SOC and AXDX testing were compared for species (ID) and AST results. Discrepant results were adjudicated using Vitek2 (Vitek MS, Marcy-l’Étoile, France) and MALDI-ToF for ID and using broth microdilution as specified by the CLSI (12) in triplicate (using the modal minimum inhibitory concentration [MIC] value as correct) for AST with technologists blinded to the results of previous testing.

Medical intervention

As part of routine clinical duties, the medical microbiologist on call (MMOC; all MMOCs were medical doctors with Royal College of Physicians and Surgeons of Canada fellowship qualification in medical microbiology) followed all BCs and intervened as indicated using her or his clinical judgment. Interventions by the MMOC were based on the SOC, patient data from the electronic medical record, and discussion with clinicians. Microbiologists were blinded to AXDX results. The significant findings list (SFL), which includes positive BCs, was printed twice daily (10:00 a.m. and 2:00 p.m.) for MMOC review. Timing of the SFL review was based on microbiologist preference, but it was always done twice daily (morning and afternoon). Patients were followed per the MMOC’s clinical judgment until microbiology laboratory reports were finalized. Any recommendation pertinent to the practice of a microbiologist was recorded by the MMOC, including antimicrobial therapies, changes in isolation precautions, investigations, and specialist consultations.

During the study, the MMOC documented treatment recommendations made for AXDX patients and the information available at the time of intervention (Gram stain, ID, or AST results). The clinical team caring for the patient continued routine medical care. Changes in antimicrobials made by the treating team in the 12 hours before the MMOC’s SFL review, as documented in the electronic medical record or through discussion with clinicians, were documented. Narrowing of antibiotics was defined as changing from a broad-spectrum antibiotic (vancomycin, piperacillin–tazobactam, carbapenem, linezolid, daptomycin, fourth- and fifth-generation cephalosporins) to narrower spectrum antibiotics, removing dual antimicrobial coverage, or changing from intravenous to oral therapy. Escalation was defined as changing from no antibiotic to any antibiotic, from a narrow-spectrum to a broad-spectrum agent (listed earlier), addition of a second drug, or changing from oral to intravenous therapy (without changing from a drug that was listed as resistant to one that was sensitive). Changing from resistant to sensitive reflected a change in antimicrobials from one for which resistance was known or predictable to one known or likely to be sensitive.

Data analysis

Laboratory results from AXDX and SOC methods were recorded in a Clindex electronic trial management system (Fortress Medical Systems, Hopkins, Minnesota). Comparison of AXDX with SOC (or adjudicated result as described earlier) provided essential agreement (i.e., MIC in a doubling dilution); categorical agreement (i.e., agreement in sensitive [S], intermediate [I], resistant [R] categorization); very major error (i.e., AXDX reported a susceptible result, and the reference method reported a resistant result); major error (i.e., AXDX reported a resistant result, and the reference method reported a susceptible result); or minor error (i.e., either AXDX or the reference method reported an intermediate result, and the other system reported a susceptible or resistant result).

ID and AST performance calculations for the AXDX Food and Drug Administration (FDA)–cleared panel were computed using a custom algorithm developed in Microsoft Visual Basic for Applications (version 7.1; Microsoft Corporation, Redmond, Washington) and output in Microsoft Excel. Because there were correctly identified species that were not officially part of the FDA-cleared panel, an adjustment for ID performance was made for any coagulase-negative Staphylococcus or Streptococcus species that was correctly reported as such on the AXDX system, despite specific species not being on the FDA-validated panel.

Results

Identification

After exclusions, 158 samples were tested (Figure 1). Four samples were excluded because of technical or administrative errors, 5 samples were excluded because of a refrigerator failure resulting in out-of-temperature-range storage of kits, and 5 failed to provide a result. One hundred forty-four samples provided a valid ID—141 specimens identified as mono-microbial by the SOC and 3 specimens identified as poly-microbial by the SOC (despite a single morphology seen on Gram stain): Klebsiella pneumoniae/Raoultella spp, Enterococcus fecalis/S. aureus, Bacteroides fragilis/Peptostreptococcus. These 3 BC specimens were identified as mono-microbial by AXDX, and the predominant organism was correctly classified (two on-panel, one off-panel). These organisms are considered correct for the remainder of the analysis.

Figure 1:

Flow diagram for samples

* Species not eligible for AST on AXDX system

† AXDX positive for SAU, SOC positive for CNS

ECO = Escherichia coli; KLE = Klebsiella spp; ENT = Enterobacter spp; PRO = Proteus spp; CIT = Citrobacter spp; SMA = Serratia marcescens; PAE = Pseudomonas aeruginosa; ABA = Acinetobacter baumannii; SAU = Staphylococcus aureus; SLU = Staphylococcus lugdunensis; CNS = coagulase-negative Staphylococcus spp; EFS = Enterococcus faecalis; EFM = Enterococcus faecium; STR = Streptococcus spp; CAL = Candida albricans; CGL = Candida glabrata; OFF = off-panel organism

The AXDX provided 94.0% sensitivity and 99.9% specificity for individual on-panel probe–target combinations for ID when only organisms validated as on-panel by the FDA were considered. When FDA-cleared and non-FDA-cleared organisms were considered, sensitivity was 95.3% and specificity was 99.9%. Specifically, of the 147 total reportable organisms (from 144 specimens), 109 organisms (75.7%) were on-panel (i.e., the FDA-cleared panel) and 11 organisms (7.6%) were correctly identified as streptococci despite not being officially part of the FDA-approved panel (5 S. salivarius, 2 S. intermedius, 1 S. oralis, 1 S. dysgalactiae, 1 S. infantis and 1 S. gordonii). In addition a S. pettenkoferi was correctly identified as coagulase-negative Staphylococcus. An additional 6 (4.1%) organisms were identified correctly as off-panel organisms for a total correct ID of 127 (86.4%). Eighteen (12.2%) organisms failed to be identified and two (1.4%) organisms were incorrectly identified: AXDX identified 1 coagulase-negative Staphylococcus as S. aureus and one off-panel organism (B. fragilis) as coagulase-negative Staphylococcus.

Time to results was consistent with the package insert, with a median of 2.1 (SD 0.65) hours for ID and 7.3 (SD 0.66) hours for AST, including hands-on laboratory time from time of positivity in the BC machine to actionable result for eligible organisms. SOC showed a median time to ID of 14 (SD 13) hours; AST showed a median of 35 (SD 12.2) hours. The difference in median time between AXDX and SOC for ID was 11.9 hours; between AST and SOC, it was 27.7 hours (Figure 2).

Figure 2:

Time to results for ID and AST comparing SOC with AXDX: median (middle line), 25th and 75th percentile (bottom and top of box), and minimum and maximum values (bars)

ID = identification; AST = antimicrobial susceptibility testing; SOC = standard of care; AXDX = Accelerate Pheno system.

Antimicrobial susceptibility testing

Susceptibility testing was available for 86 of 144 specimens (59.7%), although 1 specimen reported AST on an incorrect ID, leaving 85 evaluated specimens (59.0%) with AST results (Figure 1). The AXDX provided 97.5% essential agreement on evaluable bug–drug combinations (discrepancies resolved by additional testing; see the Methods section). There were no very major errors, 3 (0.5%) major errors (a K. pneumoniae with piperacillin–tazobactam, a K. pneumoniae with cefazolin, and an E. coli with ceftazidime); and 19 minor errors. For the one coagulase-negative Staphylococcus misidentified as S. aureus, despite the differences in breakpoints for different staphylococcal species, the results were in essential agreement and categorical agreement except for erythromycin, which showed a major error.

Clinical intervention

Of 144 reportable specimens, 46 (31.9%) had no microbiologist intervention or change in antimicrobials by the clinical team. Overall, there were 113 interventions on 100 specimens for an average of 0.77 interventions per specimen and 1.13 interventions per intervened specimen. The majority (62; 54.9%) of interventions involved narrowing antibiotics. These changes were made when the following information was available: Gram stain, 32 (28.3%); ID, 38 (33.6%); or AST, 43 (38.0%). The ward team made 67 (59.3%) interventions, most frequently after the Gram stain results (22; 32.8%) or AST results (27; 40.3%) were available. Most interventions made by the clinical team after Gram stain (19 of 22; 86.4%) escalated coverage (usually the addition of an antibiotic when none had been started), and most interventions made with AST information (25 of 27; 92.6%) resulted in narrowing of antimicrobial coverage (Figure 3). Microbiologists made 46 interventions; 10 at Gram stain, 20 at organism ID, and 16 when AST was available. The availability of Gram stain allowed recommendations that were evenly distributed between narrowing, escalating, investigating, and consulting (Figure 3). At the time of ID, 11 of 20 (55.0%) recommendations were to narrow, 5 of 20 (25.0%) were to escalate, and the remainder were for investigations or consultations. Finally, when AST was available, the majority of recommendations for intervened patients (11 of 16; 68.8%) were narrowed.

Figure 3:

Interventions performed by the clinical team (team) or the MMOC

Narrow = reducing antimicrobial spectrum (see text), escalate increasing antimicrobial spectrum (see text); investigation = MMOC suggests further testing (e.g., radiology, culture); suggest consult = consultation with another service recommended; change R to S = recommendation to change from an antibiotic to which the organism is resistant to one to which it is susceptible; MMOC = medical microbiologist on call.

The median time to microbiologist intervention from BC positivity was 26.1 (SD 10.9) hours for ID and 46 (SD 17.7) hours for AST. The earliest potential time for intervention using SOC would have been at a median of 14 hours (SD 13) for ID and 35 (SD 12) hours for AST, respectively, assuming that intervention was made as soon as results were available. If AXDX had provided ID and AST results, the median time to intervention could have been reduced to 2.1 (SD 0.65) hours for ID and 7.3 (SD 0.66) hours for AST (Figure 2). Comparing current workflow with AXDX data, including the delay between result availability and SFL review by the MMOC, there would have been a 24.0-hour improvement in interventions based on ID results and a 38.7-hour improvement in interventions based on AST results. Using best possible time (assuming the MMOC intervened immediately when information was available from the SOC), the difference would be 11.9 hours and 27.7 hours for ID and AST, respectively.

Discussion

We examined the potential utility of AXDX as part of bacteremia management. The AXDX technology worked according to the manufacturer’s claims and other centres’ experiences (13,14), providing ID and AST substantially faster than the SOC. We chose to test only BCs with Gram stain results that would likely be on-panel for the AXDX, namely those with a single organism appearing to resemble the Enterobacteriaceae, Staphylococcus, Streptococcus, or oval yeast. Despite this, there were still off-panel organisms, and two specimens were poly-microbial. For the poly-microbial cultures, AXDX identified the predominant organisms, suggesting the second bacteria may have been present at a lower concentration. These results are consistent with previous studies examining BCs for Gram-negative pathogens (15). It seems reasonable, therefore, that using Gram stain as a selection criterion could increase cost effectiveness of the AXDX platform.

The primary benefit for fast ID and AST is improved time-to-optimized antimicrobial therapy. In a low-resistance environment, empiric therapy is often much broader than ultimately necessary. This presumption is supported by the fact that most interventions narrowed antimicrobials. Given that the risks of broad therapy increase over time (9), rapid change to narrow therapy should improve patient outcomes, reduce costs, and minimize resistance. On the basis of our data, it could be possible to narrow therapy as much as 38 hours earlier if rapid diagnostic testing was part of a stewardship intervention that included real-time interventions based on rapid testing results. In our context, there was little escalation of therapy and only two changes from a drug for which resistance was identified to a more effective drug. In contexts in which there is more resistance or higher rates of pathogens requiring specific therapy (e.g., Stenotrophomonas), escalating antibiotics on the basis of rapid diagnostic results may have a higher impact on clinical care (16).

Although accuracy of AXDX was within acceptable limits, there were two ID errors, both of which might have affected patient care. The first patient was being treated with ceftriaxone and azithromycin for community-acquired pneumonia; an ID of S. aureus rather than coagulase-negative Staphylococcus would most likely have triggered further investigation and possibly alteration of antibiotic therapy. However, the incorrect susceptibility for erythromycin would not have been reported and thus would not have had a clinical impact. For the second patient, for whom AXDX reported coagulase-negative Staphylococcus rather than the correct organism, B. fragilis, the mismatch of the Gram stain and ID results would have been flagged at the laboratory and most likely resulted in the rejection of the laboratory report until further clarification could be sought. Nevertheless, if reported, this may have resulted in removal of anaerobic coverage for the organism that ultimately grew. Any adverse events could have been mitigated by use of Gram stain and subculture of the organism to confirm AXDX results.

It is tempting to think that faster diagnosis will mean faster changes in therapy. However, it does not necessarily follow that all of the interventions made in this study would be possible if AXDX was the primary diagnostic methodology. With the SOC, there was more time for information from diagnostic testing and patient trajectory to accumulate. Evidence of improvement and more secure diagnosis may have resulted in more comfort with accepting recommendations for antimicrobial narrowing. Moreover, although 24-hour AXDX results reporting could be implemented, if the clinicians covering at night are not familiar with the patient’s case (e.g., because of cross-coverage), they may defer optimizing antimicrobials until the clinical team is available the following morning.

Another consideration in evaluating these data is the fact that clinical decisions made by medical microbiologists are not standardized but are based on clinical judgment. However, because this evaluation was done with all the microbiologists at Vancouver General Hospital (six in total), it is a realistic reflection of current practice of multiple physicians in a major Canadian centre.

To understand the potential impact of AXDX, we assumed that interventions would be made when results were available from AXDX. This assumption is not necessarily a reflection of reality, and it is not the current workflow. A more fair comparison of the technology would be with a workflow in which microbiologists make immediate interventions at the time information is available from the SOC. For that reason, we have included two comparisons, the actual intervention and the earliest possible intervention, to provide readers with the ability to judge different workflows compared with AXDX technology. However, regardless of current workflow, to maximize clinical and fiscal benefits from AXDX, or any other rapid technology, it should be implemented as part of a package of antimicrobial stewardship interventions, including both direct communication with the clinical team and timely patient follow-up (17). The value added of the technology needs to be considered in light of what services are available to guide prescribers as soon as results are available.

Notwithstanding the limitations noted here, we would expect AXDX to have an impact on prescribing: recently presented data, from a different centre, show a mean decrease in antibiotic use by 2 days and in carbapenem use by 2.9 days, with a 3-day reduction in length of stay (18).

In summary, AXDX is a new technology that allows faster results for positive BCs with acceptable accuracy and provides the opportunity to introduce antimicrobial stewardship interventions with accurate results more than a day earlier. Its role merits further investigation with assessment of impact when results are available for action and to assess the economic and clinical benefits.

Competing Interests:

None declared.

Ethics Approval:

N/A

Informed Consent:

N/A

Registry and the Registration No. of the Study/Trial:

N/A

Funding:

No funding was received for this work.

Disclosures:

The authors were provided with reagents and use of the machine by Accelerate Pheno for the purposes of the study. The authors are solely responsible for the design of the study and its content.

Peer Review:

This article has been peer reviewed.

Animal Studies:

N/A