Discussion

In this retrospective cohort study of blood cultures identified via BCID compared to those identified via standard culture-based techniques, BCID use was associated with a reduction in time to organism identification which allowed earlier initiation of optimal therapy and de-escalation of anti-MRSA agents, and a trend toward earlier anti-PsA agent de-escalation. Lack of ASP availability overnight did not significantly impact the primary outcomes. Vancomycin DOT for CoNS was numerically lower with BCID use, however this finding was not significant. These results indicate that rapid pathogen identification diagnostics allow for more optimized antimicrobial use even when implementation is restricted to daytime hours.

The magnitude of the reduction in time to optimal therapy (28.7 h) is less compared to Chiasson et al, which found a reduction in time to optimal therapy of 39.1 hours with BCID implementation (73.8 h vs 34.7 h). ^{Reference Chiasson, Smith, Jodlowski, Kouma and Cutrell3} That study defined optimal therapy as the preferred antimicrobial agent based on evidence-based consensus guidelines, which may be more broad than our definition in which more narrow-spectrum antimicrobials were preferred. Additionally, Chiasson et al excluded polymicrobial blood cultures which may have made the studied patient population less complex to manage. Our observed reduction in time to anti-PsA agent de-escalation (16 h) is similar to previously reported reductions in piperacillin-tazobactam use of 11–12 hours. ^{Reference Banerjee, Teng and Cunningham4} Although no significant reduction in vancomycin DOT for CoNS was detected, a numerical reduction from 4 to 3 days may be of clinical relevance.

To date, no study has demonstrated a difference in mortality associated with the use of BCID. Most critically ill patients empirically receive vancomycin in combination with an anti-pseudomonal beta-lactam meaning that rapid diagnostic testing is likely to impact mortality only when fungi or highly resistant gram-negative bacteria are detected, such as ESBL-producers or carbapenem-resistant Enterobacterales. Our study did not capture any infections caused by carbapenem-resistant Enterobacterales, and only 13 total cases of infections caused by ESBL-producers or fungi. It is therefore likely that these pathogens did not cause infection with enough frequency to impact observed mortality in this study.

Unlike previous studies, this analysis failed to demonstrate a benefit on hospital length of stay. This may be attributed to the presence of untested effect measure modifiers such as social barriers to hospital discharge (eg insurance status, lack of caregiver support) which are common at our institution, and/or the inclusion of patients with polymicrobial bloodstream infections. There were numerically more cultures with polymicrobial growth in the BCID group, which may have delayed optimal therapy or affected hospital length of stay. The median hospital length of stay for those with polymicrobial growth was 25.4 days, numerically longer than the median duration of hospital stay for both the BCID and standard groups (17.4 and 14.4 d respectively).

Use of antimicrobials targeted at another site of infection was expected to affect the ability to de-escalate from broad-spectrum antimicrobials, prevent switching to optimal therapy, and prolong hospital length of stay. In a subgroup analysis after excluding these patients, we found similar results as the primary analysis with respect to time to optimal therapy, time to anti-MRSA agent de-escalation, and all secondary outcomes. Additionally, the subgroup analysis showed a statistically significant reduction in time to anti-PsA agent de-escalation. Reduced use of broad-spectrum antimicrobials has other benefits not quantified in this study, including decreased risk of Clostridioides difficile infection, development of fungal and antimicrobial-resistant bacterial infections, and vancomycin associated acute kidney injury. ^{Reference Sinha Ray, Haikal, Hammoud and Yu7–Reference Hawkins, Raux, Weeda and Hornback13}

Infectious diseases consultation is a potential confounder that was not assessed in this study. Expert consultation could result in quicker selection of optimal antimicrobial therapy, increased likelihood of source identification and control (ie central line removal), improved mortality, and shortened length of hospital stay. ^{Reference Kobayashi, Marra and Schweizer14–Reference Tholany, Kobayashi, Marra, Schweizer, Samuelson and Suzuki17} While not a replacement for infectious diseases consultation, ASP support does provide expert guidance on antimicrobial use. Our analysis showed that BCID use was still associated with decreased time to optimal therapy and time to de-escalation of MRSA/PsA agents when ASP support was not available. This suggests successful uptake of BCID into providers’ practice patterns despite a limited implementation schedule. Additional confounders may have been present at hours when BCID was not available (eg reduced staffing and differing clinician practices) and represent limitations of this natural experiment approach.

The ability to de-escalate based on BCID results requires confidence that these results are accurate. Thus, the eleven cases in which the BCID results were discordant with standard culture-based techniques are of interest (Table S2). Discordant cases generally fall into one of four categories: (1) failure to detect an organism on the panel (five instances), (2) failure to identify a resistance gene for an organism that was phenotypically resistant (two instances), (3) inability to differentiate between CoNS species (three instances), and (4) detection of an organism that was not identified via standard techniques (one instance).

Situations (1) and (2) represent scenarios in which failures are likely to have more severe consequences. Targeting therapy based on incorrect results in these cases results in a lack of coverage of the causative pathogen, possibly during the first 48 hours of a critical illness. Four of the five cases of BCID failing to detect an organism occurred in polymicrobial cultures. Discordance in three of these cases is attributed to growth in the opposite bottle within the set (eg aerobic bottle grew Corynebacterium, while the anaerobic bottle grew B. fragilis; BCID was run only on the aerobic bottle) and therefore not representing a true failure of the BCID, but rather a shortcoming of BCID being run on only one bottle. Of the two instances in which resistance gene results incorrectly predicted the isolate’s susceptibility, one involved an isolate of MRSA in which neither mecA nor mecC were detected. In this case, the BCID was repeated on a different bottle also growing MRSA, which again was negative for mecA and mecC. However, three days later, methicillin-sensitive Staphylococcus aureus (MSSA) was isolated and the consulting ID team felt this patient who used injection drugs likely had both MRSA and MSSA bacteremia. The other involved detection of mecA from a polymicrobial culture containing two CoNS in which both isolates were susceptible to oxacillin. Situations (3) and (4) represent scenarios in which failures are less likely to have severe consequences, resulting in unnecessary use of broad-spectrum antimicrobials or inappropriate coverage of an organism that does not need to be targeted.

This study had several limitations including the short duration and resultant small number of cultures assessed. At our institution, BCID panels are run on only the first positive bottle in a blood culture set, unless a second bottle reveals organisms with a different Gram stain morphology. This seldom resulted in organisms, most frequently suspected contaminants, being missed from BCID not being run on multiple bottles; however, these cases were still included based on correctly identifying what grew in the bottle sampled. Additionally, the underlying source of the bloodstream infection and the ability for source control were not assessed, which may affect duration of therapy and hospital length of stay. Strengths of the study include the use of a contemporaneous control group, individual chart review by at least two study investigators, and inclusion of real-world complex patients (ie polymicrobial bloodstream infections).

In summary, use of BCID was associated with significant reductions in median time to optimal therapy and median time to de-escalation of anti-MRSA agents. Additionally, BCID use was associated with a significant reduction in median time to anti-PsA agent de-escalation when controlled for antibiotics targeted at other infection sites, which has not been previously studied. These improvements were realized despite BCID being available only during daytime hours and lack of access to 24/7 ASP support. This finding has implications for the adoption of novel diagnostic technologies in other settings with limited personnel resources.