Discussion

We retrospectively measured AU and AU/AR ratios facility-wide and in select ICUs in two hospitals from 2020 through 2022 and compared AU and AU/AR values between hospitals. We found that accounting for the number of AR events reduced the number of statistically significant differences between hospitals. However, when members of antimicrobial stewardship committees assessed AU and AU/AR values, more respondents were able to assess antimicrobial use and identify potential antimicrobial overuse based on AU/AR values than AU values alone. This suggests that benchmarking select AU to AR events provides potentially actionable metrics that stewardship programs can use to prompt audits of antimicrobial administration to determine appropriateness.

We found most strikingly that echinocandins were very frequently used relative to FRC AR events in Hospital A. This contrast led the AU/AR ratio for this pair to be very high, leading most antimicrobial stewardship committee members to indicate potential overuse. This should prompt a review of indications for echinocandin use, and chart audits to determine appropriateness of use. Indications for echinocandin use include treatment of candidemia,^{Reference Moreno-Garcia, Puerta-Alcalde and Gariup18} chronic disseminated candidiasis,^{Reference Cornely, Bangard and Jaspers19} Candida peritonitis,^{Reference Carneiro, Mavrakis and Mylonakis20} esophageal candidiasis,^{Reference Kartsonis, DiNubile, Bartizal, Hicks, Ryan and Sable21} Candida osteoarticular infections,^{Reference Gamaletsou, Kontoyiannis and Sipsas22} invasive aspergillosis as salvage therapy,^{Reference Aruanno, Glampedakis and Lamoth23} and prophylaxis against invasive fungal infections in hematopoietic cell transplant recipients.^{Reference Epstein, Seo, Brown and Papanicolaou24} Notably, suspected infections from Candida and Aspergillus may not be microbiologically confirmed with culture due to inability to sample infected sites or inadequate sensitivity of microbiological cultures in identifying the organism. Moreover, cultures may not always be indicated, as in the case of esophageal candidiasis where the diagnosis is frequently made clinically. Antimicrobial stewardship committees in partnership with prescribers need to define indications for echinocandin use that encompasses both straightforward culture-proven infections, and more nuanced indications that are not microbiologically-proven.

We also found that IV vancomycin was frequently used relative to MRSA AR events in Hospital A and B. Indications for IV vancomycin use include treatment of endocarditis caused by corynebacteria, enterococci, staphylococci, and streptococci;^{Reference McDonald, Aggrey and Tarik Aslan25} staphylococcal infections (eg bloodstream infections, bone infections, lower respiratory tract infections, skin and skin structure infections);^{Reference Liu, Bayer and Cosgrove26} central nervous system infections (eg brain abscess, epidural abscess, bacterial meningitis, cerebrospinal fluid shunt infection);^{Reference Tunkel, Hasbun and Bhimraj27} endophthalmitis;^{Reference Durand28} peritonitis in the setting of peritoneal dialysis;^{Reference Li, Chow and Cho29} and prosthetic joint infection.^{Reference Osmon, Berbari and Berendt30} IV vancomycin is also indicated for surgical prophylaxis of patients at high risk for MRSA infection.^{Reference Bratzler, Dellinger and Olsen31} Infections for which IV vancomycin are indicated are sometimes not associated with positive cultures, as with some bone and joint infections, pneumonia, cellulitis, central nervous system infections, and prosthetic joint infections. Moreover, IV vancomycin is indicated for empiric coverage of MRSA or methicillin-resistant S. epidermidis. Antimicrobial stewardship committees need to define indications for IV vancomycin use for which its use is appropriate. Nevertheless, the IV vancomycin/MRSA ratio can be used by antimicrobial stewardship teams in educating prescribers about the relative paucity of MRSA events compared to the degree of vancomycin use, and can be used to measure the effects of interventions to promote more judicious use.

Linezolid was frequently used relative to VRE AR events in Hospital B facility-wide, leading the majority of respondents to indicate potential overuse. Linezolid AU was also significantly greater in Hospital B compared to Hospital A. These results should prompt an examination of linezolid AU in Hospital B to determine appropriateness of use.

Appropriate linezolid use includes treatment of VRE,^{Reference Birmingham, Rayner, Meagher, Flavin, Batts and Schentag32} pneumonia caused by Streptococcus pneumoniae and S. aureus,^{Reference Wunderink, Niederman and Kollef33} skin and skin structure infections caused by S. aureus, S. pyogenes, or S. agalactiae,^{Reference Wilcox, Tack and Bouza34} drug-resistant tuberculosis,^{Reference Conradie, Bagdasaryan and Borisov35} select non-tuberculous mycobacteria,^{Reference Winthrop, Ku and Marras36} and nocardiosis.^{Reference Davidson, Grigg, McGuinness, Baird and Anstey37} Notably, these conditions may not be associated with positive cultures as in the case of pneumonia or skin and skin structure infections. Indications for linezolid need to be formalized by antimicrobial stewardship committees to inform audit and feedback efforts.

Carbapenems were frequently used relative to ESCR AR events in the medical ICU of Hospital A and the ICU of Hospital B, leading the majority of respondents to indicate potential overuse of carbapenems in these settings. Indications for carbapenems include sepsis and septic shock,^{Reference Roberts, Kirkpatrick, Roberts, Robertson, Dalley and Lipman38} complicated urinary tract infection,^{Reference Ezure, Rico and Paterson39} intra-abdominal infection,^{Reference Zilberberg, Nathanson, Ditch, Lawrence, Olesky and Shorr40} healthcare-associated or high-risk community-acquired infection,^{Reference Harris, Tambyah and Lye10} bacterial meningitis,^{Reference Baldwin, Lyseng-Williamson and Keam41} and moderate to severe skin and skin structure infection.^{Reference Fabian, File and Embil42} These conditions may or may not be associated with positive cultures, and use within these indications would be considered appropriate. However, as antimicrobials reserved for the treatment of antibiotic-resistant Gram-negative organisms such as ESCR Enterobacterales,^{Reference Harris, Tambyah and Lye10} appropriateness of carbapenem use can be further classified as appropriate but suboptimal versus appropriate and optimal. Appropriate but suboptimal use can be defined as carbapenem administration when more narrow-spectrum antimicrobials would be sufficient based on pre-test probabilities for antimicrobial resistance in the case of empiric therapy, or culture results in the case of definitive therapy.^{Reference Spivak, Cosgrove and Srinivasan43} Antimicrobial stewardship teams need to compile a list of indications for which carbapenem use is appropriate, and develop criteria for appropriate but suboptimal carbapenem use, versus appropriate and optimal carbapenem use.

Restricted antimicrobial authorization seems to contribute to the numerically lower facility-wide AU/AR ratios for carbapenems/ESCR and new beta-lactam beta-lactamase combination antibiotics/MDRPA plus CRE in Hospital A. However, the numerically higher facility-wide AU/AR ratios for echinocandins/FRC in Hospital A suggest that there are limits to the effectiveness of antimicrobial restriction and that audits are needed to guide further stewardship efforts. In Hospital B, restricted antimicrobial authorization also likely contributes to the numerically lower facility-wide AU/AR ratio for beta-lactamase combination antibiotics/MDRPA plus CRE. The numerically higher AU/AR ratios for linezolid/VRE in Hospital B, and IV vancomycin/MRSA in Hospital A and B, may be due in part to the lack of restrictions on ordering and administration.

The strengths of our study include use of AU and AR metrics already reported to NHSN, benchmarking of AU to AR that seems more instructive and actionable than AU metrics in isolation, and comparison of AU and AU/AR values in a university hospital and a community hospital. It, however, has limitations. First, AR events are limited to select drug-resistant microbes, which restricts the number of computable AU/AR ratios. However, this paper provides a foundation that can inform the development of other AU/AR ratios for a greater number of antimicrobials, and highlights the value of adjusting AU for AR when comparing AU across hospitals. Second, we cannot definitively assess appropriateness of AU based on AU/AR ratios alone. Indications for each antimicrobial need to be compiled by antimicrobial stewardship committees, and audits of AU need to be performed to determine appropriateness of use. However, numerically higher AU/AR values may indicate an excess of empiric coverage and motivate the performance of audits. Third, AR events are assigned to the location where samples are collected, not necessarily where patients are treated. For example, AR events identified from samples collected in the emergency department would be assigned to that department, not the inpatient unit that would provide antimicrobial therapy for the patient after admission, which would result in inflation of AU/AR ratios in receiving units. However, facility-wide AU/AR ratios would be less susceptible to this inflation and can still be useful in identifying potential excesses of antimicrobial use. Fourth, AR events can be rare or absent, resulting in widely fluctuating or non-computable AU/AR ratios, especially at the location level. However, when computable, AU/AR ratios seem more informative than AU metrics in isolation. Fifth, this was a two-center study in a single health system and its findings may not be generalizable. However, since both AU and AR metrics used for this study are already reported to NHSN, our proposed metric can be replicated in other hospitals for multicenter studies comparing AU/AR ratios. Sixth, we did not account for COVID surges in this paper since it was beyond its scope. We have a separate paper using a different dataset wherein we found that AU for broad-spectrum antibacterial agents predominantly used for hospital-onset infections was greater in COVID patients compared to non-COVID patients from March to December 2020.^{Reference Santos, Tseng and Martinez44} This analysis was not possible with the current paper since NHSN does not stratify AU by COVID status.

In summary, we compared select conventional AU metrics to AU/AR ratios in two hospitals and found that antimicrobial stewardship committee members were more able to assess antimicrobial use and identify potential overuse based on AU/AR values compared to AU values alone. Future directions in our hospitals would include updating AU/AR ratios yearly and calculating it facility-wide and for each inpatient unit. AU/AR ratios can be readily calculated by facilities that already monitor AU and AR events using NHSN methods. It can be used to benchmark AU for interhospital comparisons and assist stewardship committees in identifying potential overuse of specific antimicrobial agents.