Discussion

Our quasi-experimental study included both before-and-after group analysis as well as a contemporaneous controlled comparative analysis of COVID versus non-COVID wards. One of this study’s objectives was to assess the potential impact of AMU for azithromycin and ceftriaxone prescribing as it pertained to the timing of the COVID admission order set implemented locally in Calgary. These agents were also frequently employed antimicrobials in other studies. ^{Reference Elligsen, Wan and Lam4,Reference Rose, Baggs and Wolford5}

There was a significantly increased rate of use seen following the implementation of the order set, concurrent with the start of the pandemic, for both antimicrobials. Aside from the azithromycin data for COVID units, this trend was not observed following the removal of both agents from the set during wave 4 in the time series analysis. The greater azithromycin use observed is likely multifactorial, including the order set, improved outcomes reported early in the pandemic based on observational data, and potentially its anti-inflammatory properties. It is also noteworthy that although COVID unit azithromycin use remained higher than baseline during wave 4, it was lower than the preceding 3 waves. Although there were consistently significant increased rates of use for both antimicrobials on COVID wards compared with control wards across all waves, the magnitude of difference was less following order-set modification, decreasing by approximately twofold for both agents. There are several potential explanations as to why this increase occurred. Early in the pandemic, there was uncertainty around the incidence of concurrent bacterial infections, and clinicians likely prescribed antimicrobials out of an abundance of caution, not wanting to undertreat and potentially cause harm to the patient, in the event there was an incipient bacterial co-infection. Even as more information became available on the low incidence of bacterial co-infection, ^{Reference Alshaikh, Godman, Sindi, Seaton and Kurdi11} the lack of effective treatment options for COVID may have contributed to higher antimicrobial prescribing rates, in addition to the prescribing inertia with antimicrobial agents empirically. It is possible that the inclusion of antimicrobials in the order set led to prescribers feeling justified in including antimicrobials for patients requiring admission to hospital for COVID.

Although the relationship is not causal, this trend of prescribing in the before-and-after analysis does illustrate the potential impact such order sets can have on prescribing behaviors, ^{Reference Zahlanie, Mang, Lin, Hynan and Prokesch20} and the importance of considering the implications of all additions made to future order sets. Other studies have demonstrated a decrease in AMU later in the pandemic, ^{Reference Cong, Stuart and AIhusein21} which is comparable to our results.

Other Canadian literature on AMU demonstrated an increase in respiratory AMU during wave 1 only on medical wards, as well as an increase in ICU AMU during waves 1 and 2 before returning to baseline during wave 3. ^{Reference Elligsen, Wan and Lam4} Our comprehensive study saw a common theme in increasing overall AMU on all wards early in the pandemic before returning to pre-pandemic levels, as well as a significant decrease in overall rates of use on COVID wards after the first wave. The exception to this was seen in azithromycin and ceftriaxone rates of use, as previously discussed. We did not specifically look at respiratory antimicrobials combined, so it is an imperfect comparison, although similar patterns exist when comparing our study to related literature. ^{Reference Elligsen, Wan and Lam4} Another noteworthy exception was seen in ICUs during wave 4 where a significantly increased rate of AMU was observed. This may have been partially driven by variant B.1.617.2 (Delta variant). This variant became the predominant strain of SARS-CoV-2 in Canada during wave 4 of the pandemic ^{Reference Chan, Irvine and Prystajecky22} and has been identified as a more virulent strain of SARS-CoV-2, causing more serious illness. ^{Reference Ong, Chiew and Ang23}

Our study also assessed AMU differences between COVID and control wards. It is interesting that for most respiratory antimicrobials, there was a consistent finding of highly significant increased rates of use on COVID wards for all waves assessed, although the magnitude of difference decreased by wave 4. Exceptions to this trend were observed with meropenem only, where greater rates of use were observed on COVID wards for wave 1 only, with no difference seen in wave 2, and greater use observed on control wards during waves 3 and 4. It may be that patients admitted to non-COVID wards more commonly had different infectious processes than their COVID-infected counterparts requiring broader spectrum antimicrobials.

There is strong evidence of increasing invasive fungal infections in critically ill patients with COVID pneumonia. ^{Reference Rovina, Koukaki and Romanou13} For instance, for COVID-associated pulmonary aspergillosis, the prevalence of this condition in critically ill patients was found to be 10% in one review and associated with a mortality rate approaching 60%, with unclear benefit of anti-mold therapy. ^{Reference Kariyawasam, Dingle, Kula, Vandermeer, Sligl and Schwartz24} COVID-associated mucormycosis, another invasive mold infection, has been shown to have a 50-fold pooled prevalence increase globally, with a mortality rate approaching 30%. ^{Reference Hussain, Riad and Singh25}

Despite these findings, our study did not demonstrate a change in antifungal use. It may be that our study was underpowered to identify a significant difference, given invasive fungal infections, despite higher incidence being described, remain a relatively uncommon diagnosis. High baseline antifungal use may also contribute to challenges in detecting a significant change, although a retrospective review by Bienvenu et al demonstrated comparable pre-pandemic ICU antifungal usage trends in Calgary. ^{Reference Bienvenu, Bestion and Pradat26}

Strengths of our study include a large data set, comprehensive study design, and complete capture of all AMU data for COVID waves included in our study. All acute care sites admitting adult patients across our health region were included. Of all quasi-experimental designs, the methodology employed has a higher internal validity. ^{Reference Harris, Lautenbach and Perencevich27} The inclusion of both a control group and pretest group strengthens the epidemiologic associations. ^{Reference Harris, Lautenbach and Perencevich27} Finally, ours is one of the first studies of which we are aware of comparing COVID and control ward AMU, with longitudinal assessments of AMU prior to and during the pandemic, which has not been commonly presented in the literature to date.

Our study was a single health region in a high-income setting ²⁸ and as such may limit generalizability to other populations. In addition, due to uncertainty around the management of COVID early in the pandemic, it would have been challenging to assess usage prospectively or in a randomized fashion given the role for empiric antimicrobials was initially unclear. Despite the limitations of our study design, we attempted to mitigate the limitations by including both pretest and control groups.

Our data was not corrected for seasonal variation which can affect AMU. Although COVID incidence may fluctuate, it is relatively seasonally indiscriminate compared with other respiratory viruses. ^{Reference Wiemken, Khan and Puzniak29} We felt this made potential confounding due to seasonality less likely.

The use of ward-level data rather than patient-level data is another potential limitation. While pharmacy records are a convenient means of obtaining a large volume of AMU data, this data is not perfect and can overrepresent AMU. For instance, an antimicrobial may have been ordered but not administered for various reasons (loss of IV access, patient refusal). Pharmacy data does not capture these events, as the antimicrobial was dispensed and therefore counted toward overall usage. Unfortunately, looking at individual patient medication administration records was not feasible given the large data set analyzed in this study. Additionally, we did not have detailed length of stay data for individual patients which can contribute to AMU. DDD per 100 patient days is an accepted measurement of incidence density to allow for comparison of antibiotic consumption.

Additionally, we evaluated various parameters including monthly dates for order-set implementation and the pandemic waves. As the exact dates do not perfectly coincide with the beginning and end of respective months, there is the possibility the data analysis for a given month may not be fully representative of the parameter being evaluated. We chose this means of evaluation as the pharmacy AMU data used in the study was provided in monthly intervals. Fortunately, the start and end dates for the parameters listed above did coincide very closely with the beginning and end of each month, which greatly reduces the likelihood of a significant impact on our analysis.

Our study demonstrated an increase in inpatient AMU assessed by DDD per 100 patient days early in the pandemic, as well as a significant decrease following the removal of antimicrobials from the Calgary COVID order set. We demonstrated a temporal association exists between order-set design and prescribing practices and the importance of considering how healthcare decision-support tool content can alter behavior. Our findings suggest that antimicrobial stewardship programs and initiatives should target decision-support tools to optimize AMU both as part of pandemic planning and outside of it. Finally, we believe our study provides a valuable addition to the COVID AMU literature. While our study cannot prove causation, our results clearly demonstrate how prescribing practices have shifted, both as SARS-CoV-2 and our understanding of it evolved over time.