Ethics

This study was conducted in accordance with national and institutional standards, as well as the Declaration of Helsinki. This study received institutional approval by the Sunnybrook Research Ethics Board. Written informed consent from study participants was obtained at the beginning of the participant surveys.

Study design and setting

This study was conducted at St. John’s Rehab (SJR) Hospital, a 178-bed rehabilitation program within Sunnybrook Health Sciences Centre, a multisite, tertiary-care, academic institution. Models of prospective audit and feedback were evaluated over 3 periods using both quantitative and qualitative methods. The baseline period spanned January 1, 2017, through December 31, 2017. The first intervention period consisted of a dedicated AMS pharmacist conducting prospective audit and feedback from January 1, 2018, to January 31, 2019. No audit and feedback occurred between February 1, 2019, and April 30, 2019. The second intervention period consisted of ward pharmacists conducting prospective audit and feedback from May 1, 2019, until April 30, 2020.

Interventions

Prior to 2017, prospective audit-and-feedback services were not provided within the rehabilitation hospital. In 2018, formal audit and feedback was provided by a trained AMS pharmacist. The pharmacist had previously worked as a full-time AMS pharmacist at the acute-care site, conducting audit and feedback among a variety of patient populations for 14 months. Twice weekly, the AMS pharmacist reviewed all patients on systemic antimicrobial therapy and reviewed cases identified as challenging with the on-call ID physician at a prearranged time, prior to recommendations being provided. Suggestions were documented in the electronic medical record, and verbal feedback was provided to prescribers by the AMS pharmacist. In addition, the AMS pharmacist delivered 4 educational sessions to nurses and prescribers on appropriate urine-culture ordering practices.

When the AMS pharmacist period ended in January of 2019 due to reductions in hospital funding, the ward pharmacists remained committed to providing ongoing antimicrobial stewardship services in addition to their regular clinical duties. This model of stewardship represents a second intervention that was implemented in February 2019. During the transition between the 2 models (February–April 2019), there was no prospective audit and feedback.

During the ward pharmacist period, 7 ward pharmacists assessed new antibiotic orders for patients under their usual clinical coverage. No formal stewardship training was provided to ward pharmacists prior to the initiation of the intervention, due limitations in resources. Throughout the entire study period, pharmacists and prescribers had access to the institution antimicrobial stewardship application, which was launched in November 2015 and included local antibiograms and dosing and treatment guidelines. Pharmacists documented their reviews in the electronic patient record, and suggestions for optimization were provided to prescribers using verbal feedback. There was no prearranged review time with an ID physician during the ward pharmacist period, but a dedicated AMS pharmacist from the acute-care hospital and an ID physician were available for consultation on an as-needed basis. Throughout both intervention periods, 13 prescribers (staff physicians) were practicing at the rehabilitation hospital.

Study outcomes

The primary outcome was the monthly total antimicrobial consumption measured by days of therapy (DOT) per 1,000 patient days (PD). Secondary outcomes included the number of urine cultures ordered monthly, the total number of suggestions, and suggestion acceptance rates. Antimicrobial stewardship suggestions included antibiotic discontinuation, optimization of dose, shortening or extending duration, and changing agent. Urine cultures were included as a diagnostic test of interest given their recognized role in driving unnecessary antibiotic prescribing, which are frequently the focus of ASP feedback. ^{Reference Leis, Rebick and Daneman12} Additional qualitative outcomes, such as the perceived successes and challenges of the ward pharmacist AMS model, were collected from the ward pharmacists using semistructured interviews. Staff physicians were surveyed to assess their perceptions of differences in audit and feedback among the models.

Data collection

Antibiotic days of therapy were extracted from the electronic patient record (MediTech). The numbers of urine cultures ordered per month were provided by the microbiology department. Bed occupancy and patient days were provided by the health records department. Patient diagnostic groups were obtained from data submitted to the National Rehabilitation Reporting System (NRS), within the Canadian Institute for Health Information (CIHI). For qualitative data collection, pharmacist interviews were conducted in person using a 5-item interview questionnaire (Appendix A). One investigator (J.C.) performed all interviews, and all answers were transcribed in written format. Physicians’ perceptions and experiences with the different models of pharmacist audit and feedback were collected anonymously through the online survey platform Qualtrics CoreXM. The survey consisted of 6 multiple-choice questions and 1 open-ended question (Appendix B). It was developed by 4 study investigators (J.C., M.E., P.L., and J.L.). Survey questions were tested by 3 of the investigators (J.L., P.L., and M.W.). Surveys were distributed to participants on July 6, 2020. A follow-up e-mail was sent on August 19, 2020, and the survey closed on August 21, 2020. No incentives were offered for participation.

Statistical analysis

An interrupted time series (ITS) analysis was conducted to analyze the outcomes of monthly DOT per 1,000 PD and monthly urine cultures ordered per 1,000 PD. In this analysis, we performed a segmented regression analysis for both measures. A level and slope-change impact model was chosen, and monthly data were modeled using a Poisson regression with the total number of patient days and urine cultures set as the offset variable in their respective models. Testing for autocorrelation and seasonality were carried out by inspecting the residuals from the autocorrelation and partial autocorrelation plots. No evidence of autocorrelation or seasonality was detected. To account for overdispersion and heteroscedasticity, robust sandwich variance estimators were used.

The proportion of pharmacist-reviewed cases for which suggestions were made and accepted during the 2 intervention periods were compared using the χ ^{Reference Dellit, Owens and McGowan2} test. Survey and interview results were analyzed descriptively. Transcribed responses from the pharmacist interview were reviewed and categorized into common themes through the identification of recurring words. All statistical analyses were performed using R version 3.4.3 statistical software (R Foundation for Statistical Computing, Vienna, Austria).