Discussion

There is very little published data exploring the real-world clinical implications of discrepancies identified in a clinical microbiology laboratory that employs a rapid commercial AST platform alongside routine susceptibility testing. ^{Reference Robinson, Stilwell and Attai11} Even with highly accurate FDA-approved platforms, occasional discrepancies between AST methods are to be expected when performing duplicate testing on the same organism. This is especially true for some broad-spectrum antibiotics such as piperacillin-tazobactam, which has been historically difficult to perform accurate and reproducible AST even by gold standard methods. ^{Reference Henderson and Humphries21} We report high concordance between the Accelerate Pheno™ and the automated Vitek®2 AST results in a real-world setting. Overall categorical and essential agreement were >90%, similar to other published data. ^{Reference De Angelis, Posteraro, Menchinelli, Liotti, Spanu and Sanguinetti15,Reference Lutgring, Bittencourt, McElvania TeKippe, Cavuoti, Hollaway and Burd22,Reference Charnot-Katsikas, Tesic and Love23} Though discrepancies were uncommon overall, they can lead to changes to antibiotic therapy, provider confusion, or potential distrust of results. Similar to previously published studies, we note that the majority of discrepancies (88.4%) tended to overcall resistance compared to the Vitek®2. ^{Reference Truong, Mongkolrattanothai, Flores and Dien Bard6,Reference Charnot-Katsikas, Tesic and Love23–Reference Lee, Scardina, Zheng and Patel25} From a patient safety perspective, rapid susceptibility methods that may overcall resistance are generally preferred to those that potentially under-estimate resistance, given the risk of under treatment in the latter.

Robinson et al (2021) reviewed all gram-negative blood cultures and compared clinical impact of discordance between standard of care to Pheno testing in a subset of cultures. They cited a higher number of patients having a prescribing impact based on discrepancies, 55%, with the most common being continuation of overly broad therapy. ^{Reference Robinson, Stilwell and Attai11} In our study, discordance leading to inappropriate changes in antimicrobials were infrequent (9%). However, conflicting results and the need to change antimicrobial therapy multiple times can lead to clinician mistrust of results in the future. Some infectious disease physicians documented their preference to “wait for final susceptibilities” before changing antibiotics. Therefore, additional education with antimicrobial stewardship groups and published discrepancy analyses are helpful to show the low incidence of discordance, the nature of common discordant results, and outline the laboratory and clinical approach to adjudication. Microbiology laboratories performing testing should maintain technical reporting guidelines for testing personnel, indicating when to suppress antibiotics, when to perform discrepancy testing, when to consult with an on-call clinical microbiologist, and how to report results. ^{Reference Truong, Mongkolrattanothai, Flores and Dien Bard6}

For example, given that ceftazidime is not used as a primary empiric agent at our institution (cefepime or ceftriaxone are favored) and the high rate of ME and mE for this antibiotic, we suppressed the results of ceftazidime from future rapid AST reporting. We felt that the “overcalling” of ceftazidime resistance, even when ceftriaxone susceptible, influenced clinicians to continue cefepime or meropenem, due to concern for the presence of an extended-spectrum beta-lactamase (ESBL), as shown in cases 1, 2, and 3 (Table 2).

Technicians were instructed to suppress and then repeat testing if Accelerate Pheno™ results did not make sense. For example, if results demonstrated intermediate or resistant results to cefepime, carbapenems, or piperacillin-tazobactam but were reported as susceptible to ceftriaxone or ceftazidime, technicians were asked to suppress these results. Oftentimes, these patterns were not noticed until after results had been released to clinicians, and discrepancies with final AST were already reported. Many laboratories are not fully staffed with clinical microbiologists 24/7, and complex manual reporting/suppression guidelines may be difficult to teach or enforce among staff who have little specialized training in microbiology. Ideally, reporting contingencies should be built into the electronic medical record or with custom rules on the diagnostic device to reduce the probability of errors.

Discrepant AST results may also have important implications for infection control practices. Note that one of the patients in Table 2 was flagged as having an ESBL, leading to changes in their isolation status per infection control, which was later redacted based on the final susceptibilities. If a rapid AST platform overcalls resistance, erroneous patient cohorting may occur occasionally. It is important for hospitals auto-flagging-resistant isolates within their electronic medical records to be aware that these discrepancies can occur and may affect patient contact precautions and isolation status.

Table 2.

Discordant results for 8 patients whose antibiotics were changed due to discrepancies in susceptibility methods. Only AST results with discrepancies that affected therapy are shown

S = susceptible, I = intermediate, R = resistant

1. ESBL; extended-spectrum beta-lactamase

This study is subject to a number of limitations. This was a single-site experience at an academic teaching hospital with a well-established antibiotic stewardship program, so results may not be applicable to all institutions. Some antibiotics included in the performance characteristic analysis were not released to clinicians as they were RUO at the time of testing (eg, Ciprofloxacin), so these discrepancies had no potential for clinical impact. Though the Vitek®2 system is one of the most commonly used AST methods in clinical microbiology labs, it is not a true broth microdilution method, instead extrapolating AST results based on growth curves. The Vitek®2 also uses a proprietary advanced expert systemReference Carvalhaes, Shortridge, Woosley, Gurung and Castanheira ²⁶ to adjust susceptibility results when unusual phenotypes are produced by testing, which could result in a categorical change, and lead to additional discrepancies. Although we used the Vitek®2 as a comparator, it is not a true gold standard, and discrepancies may have been due to errors in either platform. Tie breaking discrepancy testing was not performed in all cases, which would have helped better characterize the nature of these discrepancies. Future comparison studies are needed to adjudicate methodological accuracy between these two platforms. Although changes in antibiotic therapy were evaluated, clinical outcomes including increased length of stay and mortality were not. Due to the nature of a small retrospective study, this would have been difficult to determine based on antibiotic selection alone.

Though AST is not reported from the Accelerate Pheno™ for polymicrobial samples, there may be instances where the Accelerate Pheno™ identifies an organism and performs susceptibilities, and an additional organism grows later in culture. In these scenarios, antibiotic narrowing may not be appropriate and could lead to undertreatment. Of note, one patient was identified where the Accelerate Pheno™ reported a Proteus mirabilis with ceftriaxone susceptibility; however, the final sub-culture grew both an E. coli and Proteus mirabilis. The E. coli was resistant to ceftriaxone. In this case, antibiotics were inappropriately narrowed in the setting of a cryptic polymicrobial infection. Polymicrobial bloodstream infections are uncommon overall, but clinical laboratories should be aware of this limitation.

Implementation of rapid susceptibility testing platforms requires antibiotic stewardship efforts to help clinicians determine the best course of action when discrepancies arise. AST reporting should be tailored to an individual institution’s needs, and laboratories may consider automated reporting rules and selective suppression for non-preferred or non-formulary antibiotics. Discrepancy testing should be performed when possible.