This study evaluated changes in antimicrobial use associated with implementation of an antimicrobial stewardship audit and feedback program using a controlled interrupted time series design . We hypothesized that implementation of audit and feedback would lead to reduced antimicrobial use in both units.
Study Setting and Population
This study was performed in four adult ICUs at St. Michael’s Hospital, a 465-bed academic teaching hospital in Toronto, Ontario, Canada. The intervention ICUs included a 19-bed trauma and neurosurgery ICU (TNICU) and a 24-bed medical and surgical ICU (MSICU). The control ICUs included a 15-bed cardiovascular surgery ICU (CVICU) and a 10-bed cardiac ICU (CICU).
Antimicrobial use and other outcomes (see below) were collected for all patients admitted to the ICUs during the study period. Approval was obtained from the Research Ethics Board at St. Michael’s Hospital. The Research Ethics Board waived the need for informed consent since the study used anonymous, aggregate, retrospective data.
The audit and feedback intervention was introduced in the TNICU on April 1, 2013 and in the MSICU on April 15, 2013. The pre-intervention and post-intervention periods were defined as April 1, 2012 to March 31, 2013 (pre-intervention) and May 1, 2013 to April 30, 2014 (post-intervention).
During the pre-intervention period, antibiotic selection was performed at the discretion of the respective ICU teams. During the post-intervention period, an infectious diseases trained pharmacist and physician reviewed all patients admitted to the intervention ICUs daily (weekdays only). Patients who remained in the ICU were reassessed every weekday until ICU discharge. Prescribed antimicrobials, as well as microbiology, laboratory and diagnostic imaging results were reviewed. During a daily, dedicated 30 minute meeting, the ICU team presented additional clinical details for each patient and the stewardship team provided recommendations on antimicrobial use to the team. Recommendations were made verbally and documented in the chart only if requested by the ICU team. The ICU team maintained prescribing autonomy. For patients followed by the infectious diseases service, recommendations were provided to the infectious diseases team, rather than the ICU team, to avoid conflicting advice. Advice was not provided on patients with cystic fibrosis (CF) as their antibiotic management was determined by a separate CF service, whose physicians have greater expertise in the management of this patient population.
This initiative was part of an Ontario-wide quality improvement project (Council of Academic Hospitals of Ontario Antimicrobial Stewardship Program in Intensive Care Units Project) to introduce audit and feedback programs into ICUs.
The primary outcome was total systemic (oral or parenteral) antimicrobial use in each ICU, measured in defined daily doses (DDD) per 1000 patient days per month [http://www.whocc.no/atc_ddd_index/]. Antimicrobial data was acquired from the pharmacy department as total grams dispensed to the unit per month (see Additional file 1). Patient days were obtained from the hospital’s administrative database.
Secondary outcome measures included the use of pre-specified antibiotic agents or classes, antimicrobial costs, antimicrobial susceptibility for Escherichia
coli and Pseudomonas
difficile infection incidence, and clinical outcomes, including monthly ICU mortality rates, ICU length of stay and 48 hour ICU readmission rates.
Antimicrobial costs were calculated as Canadian dollars per patient day per month and were obtained from the pharmacy database. The number and antimicrobial susceptibility of P. aeruginosa and E. coli isolates from clinical samples were assessed. These organisms were selected a priori since they were the two most commonly isolated Gram negative organisms in our intervention ICUs. Only the first isolate per patient per hospital stay was included, unless there was a change in antimicrobial susceptibility. In this case, subsequent isolates with additional antimicrobial resistance were also included. Specimens were accepted from all clinical sites cultured with two exceptions. Respiratory specimens from patients with cystic fibrosis were excluded since these patients are often chronically colonized with multi-drug resistant organisms that, in most instances, reflect antimicrobial use prior to arrival in the ICU. Additionally, screening swabs collected for infection control purposes were not included. Susceptibility data was obtained from the clinical microbiology laboratory information system.
Incidence rates of nosocomial C. difficile infection were calculated based on prospective surveillance conducted by Infection Prevention and Control. Clinical outcomes, including ICU mortality rates, ICU length of stay and 48 hour ICU readmission rates, were available via the Critical Care Information System (CCIS) [http://www.health.gov.on.ca/en/pro/programs/criticalcare/ccis.aspx].
In addition to the above outcomes, age, sex, admitting diagnosis, ventilator utilization ratio (calculated as ventilator days divided by patient days) and mean multiple organ dysfunction score were obtained using data from the CCIS. Other factors likely to influence antimicrobial use, including monthly rates of febrile respiratory illness and influenza, were also collected. Finally, data related to cystic fibrosis was documented. St. Michael’s Hospital has the largest adult CF program in North America [http://www.stmichaelshospital.com/programs/cysticfibrosis/index.php]. Patients with CF frequently receive prolonged durations of multiple, broad spectrum antimicrobials at high doses. Therefore, data collection included the number of patient days per month in each unit attributable to patients with cystic fibrosis (through International Classification of Diseases 10th version - ICD-10 codes) as this was a potential confounder with respect to overall antimicrobial use.
The CVICU and CICU served as control ICUs because these units did not receive the intervention. There was minimal overlap between attending physicians in control and intervention units. H2 blocker and proton pump inhibitors, measured in DDD per 1000 patient days, were used as negative tracer medications, since prescription of these agents should not have been affected by the intervention.
The primary outcome was assessed by segmented regression analysis of interrupted time series data . This method estimates changes in the level and trend for the outcome (i.e. antimicrobial use) after the intervention while controlling for pre-existing trends and temporal confounders. The analysis was performed separately for each of the intervention and control ICUs as well as for each of the tracer medications.
Traditional sample size calculations are not appropriate for time series analysis. Instead, it is recommended that there are a minimum of 12 data points before the intervention and 12 data points afterwards as in our study . Autocorrelation was assessed by computing the Durbin-Watson statistic. Since evidence of autocorrelation was detected, all analyses were performed using autoregression in SAS (Version 9.4, Cary, North Carolina) with correction for first and second order autocorrelation using the maximum likelihood method. The assumptions of normality, homoscedasticity, and linearity were assessed using the Q-Q plot of residuals, plot of residuals against predicted values and plots of residuals against each variable in the regression model respectively. This same method was used to assess changes in tracer medications.
Categorical variables were assessed using the Chi-square test or Fisher’s exact test, continuous variables were assessed using the t-test or Wilcoxon rank sum test, and rates were assessed using incidence rate ratios.
All tests of significance were two-tailed and a p-value less than 0.05 was considered statistically significant. For the analyses of specific antibiotic agents and classes and for the analyses of resistance of organisms, Bonferroni corrections were used to correct for multiple hypothesis testing; for the classes of antimicrobials and individual antibiotics, a p-value of < 0.0028 was considered statistically significant, and for resistance tests, a p-value of < 0.0033 was considered significant. Statistical analysis was performed using SAS (Version 9.4, Cary, North Carolina) with the exception of incidence rate ratios, where Stata (Version 13, College Station, Texas) was used.