We have implemented AXDX for rapid identification and AST, in a setting that included structured ID treatment protocols, active reporting of GNB to ID consultant and prospective audit and feedback. In a prospective evaluation of 46 and 57 patients with GNB and high risk for AMR, before and after the implementation of AXDX, respectively, we show a significant shorter time to a beneficial antimicrobial change, through escalation to an appropriate treatment, or, more commonly, de-escalation of a broad-spectrum treatment for a susceptible organism.
The merit of a novel rapid diagnostic method should be evaluated in various aspects, including analytical performance, the ability to shorten TAT in a meaningful way, and the impact on patient care.
The analytical performance of AXDX was evaluated in multiple studies. Correct microbial identification was obtained in 75.3-88.7% of bacteremias, and in 89.8-97.1% of bacteremias included on the system's panel. AST results showed an overall category agreement of 92-97.9% and 92.7%-95.4% for GNB compared with culture-based AST (3,4,16,17,5–11,15). Our experience was similar with very few major and very major errors.
Nevertheless, AXDX performed poorly analyzing cultures with polymicrobial growth. None of six polymicrobial cultures (11% of cases) were correctly identified, either because the AXDX did not identify one of two organisms (5 cases) or because two culture bottles grew different organisms (1 case). In 2 of these cases the incorrect identification by AXDX led to an inappropriate treatment de-escalation, potentially compromising patient care. Low AXDX sensitivity in polymicrobial bacteremias was found in other studies. A study reported correct identification of all bacteria of mixed cultures in 12.5-40% of cases (2,3), and in 68.8% when only bacteria included in the panel were considered (3). In another study in which sterile site specimens were inoculated in blood culture bottles, 7/9 polymicrobial cultures were wrongly identified as monomicrobial (24). In 6 studies investigating the clinical impact of AXDX implementation, 124 of 1444 cultures (8.6%) were found to be polymicrobial (11–13,18–20). In all these studies, polymicrobial cultures were excluded from the performance and impact analyses, although AXDX failure might have had a negative impact on patient care.
A significant shortening of TAT times was previously shown after AXDX implementation (2,3,15–21,4,5,9–14). In studies assessing AXDX clinical impact, the time to identification and AST was shorter than the SOC by 11.3-40.7 hours (12–14,18–21). In our study the AST results time was only 16.8 hours shorter, probably because the laboratory workflow supported performing and reading AST results during evening times.
Following the implementation of AXDX, the median time to an AST-oriented beneficial change was 20.4 hours shorter. This finding is with agreement with other studies showing a decrease in the time to laboratory-guided antimicrobial changes and down-escalation of broad-spectrum antimicrobials in various clinical settings (11–14,18,19,21,22). Of note, in all of these studies ASP was employed. In some studies (11,18,21), the time to appropriate treatment with AXDX was similar to that with SOC , as most patients received appropriate empiric treatment before culture results.
We did not find a difference between the study periods in robust clinical outcomes, including mortality, LOS, re-hospitalization or CDI incidence, and antimicrobial treatment duration was significantly longer during the intervention period. This is in agreement with a few previous studies that also did not show any impact on such outcomes (11,12,14,19). In contrary, Dare et al. showed a 1-day shorter LOS post positive culture identification (3 days for GNB), 1-day shorter duration of treatment and lower utilization of broad-spectrum antimicrobials (13). Sheth et al. showed a 1-day shorter LOS and a shorter duration of broad spectrum treatment (22). Bhalodi et al reported a 1-day shorter LOS only for patients with GNB (14). The most striking clinical impact was reported by Babowicz et al. with 85% lower risk of death after implementing AXDX (18). As there was no difference between the intervention and control groups in the time to appropriate antimicrobial treatment, the reduced mortality could only be attributed to early treatment de-escalation. Indeed, unnecessary broad-spectrum empiric treatment was associated with a 1.22-fold higher risk of death in a large observational study (25). Nevertheless, the reduced mortality reported in (18) seems to exceed the expected possible effect of antimicrobial de-escalation.
The impact of AXDX and other rapid AST technologies on patient care is complex. In order for an earlier AST result to lead to an earlier treatment change, clinicians need to be informed of the results, understand them, have confidence in the system and feel comfortable with de-escalation of treatment early on in a bacteremic patient course, when clinical signs of improvement may not yet be apparent (26). In a study by Lee et al., in cases where AXDX revealed an inappropriate empiric treatment, escalation followed in most cases. In contrast, more than half of the opportunities to de-escalate treatment were missed (20).
ASP teams can facilitate antimicrobial changes in accordance with AST results based on AXDX analysis. However, early involvement of these teams in patient care might lead to a better initial appropriateness of empiric treatment, decreasing the potential benefit of rapid AST. For example, Walsh et al. reported shorter treatment duration and LOS after the implementation of the AXDX, however the intervention also included reporting of cases of bacteremia to the ASP team, which led to a significant increase in the involvement of ID consulting the management of bacteremic patients (21).
Rapid identification and AST using various laboratory methods were shown to increase cost effectiveness in the US healthcare system. None of these studies have included analyses of AXDX (27,28). Concurring with other studies, we also show the main benefit of AXDX in rapid antimicrobial de-escalation. Future studies are needed to determine whether AXDX is cost effective in terms of ASP.
The strengths of this study include the similar use of laboratory reporting and ASP team involvement in patients' care before and after AXDX implementation, which isolates the benefit of AXDX to its laboratory performance only; the inclusion of cases with a higher rate of AMR, who might have more benefit from rapid AST. In contrast to previous studies that excluded cases in which AXDX failed to identify bacteria (such as technical failures, incorrect identification of polymicrobial bacteremias etc.), we have included these cases, hence data was analyzed according to an "intention-to-diagnose" design. The study has some important limitations. As it was performed in a single and relatively small hospital, its results might not be generalizable to other settings, where larger volumes of cultures are processed, and early and intensive involvement of ASP clinicians in the management of bacteremic patients might not be feasible. The definitions of escalation and de-escalation are debatable in some cases, e.g., when switching between broad-spectrum antimicrobials such as third generation cephalosporins, quinolones, aminoglycosides etc. We have used the approach that directs ASP decisions in our institute. Last, due to the relatively small number of cases, our study was under-powered to detect differences in outcomes such as mortality and LOS.