2.1 Setting and design
This study was conducted in Centro Hospitalar Universitário de São João, a 1100-bed tertiary care public teaching hospital in Porto, Portugal. A set of restrictive AMS strategies including formulary restriction, pre-authorization of restricted antimicrobials and automatic stop-orders have been implemented and sustained in all wards for several years. Restricted antimicrobials included carbapenems, quinolones, colistin, daptomycin, linezolid among others. Pre-authorization was performed by the AMS team since 2014 and all restricted antimicrobials were reviewed in a time frame no longer than 24 hours in week days. In 2014, following the national legislation, the Infection and Antimicrobial Resistance Control and Prevention Unit (UPCIRA) was created and became responsible for both infection prevention and control and AMS program. Since then, UPCIRA has implemented several persuasive AMS interventions, particularly prospective audit and feedback, in various wards, including Orthopaedics, Plastic Surgery and Burn Unit, Cardiac Surgery and Urology. In April 2016, an outbreak of carbapenemase producing Klebsiella pneumoniae was identified in the Vascular Surgery ward. Despite having been identified in April 2016, the first case of the outbreak was traced back to January 2016. The identification of this outbreak prompted UPCIRA to act and alongside with infection control and prevention measures, an analysis of the factors that might have contributed to the occurrence of the outbreak identified excessive use of antibiotics and a high proportion of patients under carbapenem therapy; also, there were no local guidelines for treatment of diabetic foot infections, one of the most common causes of admission to Vascular Surgery ward. Prospective audit and feedback of antimicrobial prescription and production of a local guideline ensued, and both were fully implemented by the end of May 2016. The outbreak ended in July 2016 and in total nine patients in Vascular Surgery ward were affected, with only two of those having infection related with KPC.
We designed a controlled ITS of the period between January 2012 and May 2018 to assess the effect of persuasive AMS strategies on the following endpoints: carbapenems consumption, total antibiotic consumption and proportion of antibiotic-free days.
The period between January 2012 and March 2016 was considered as the pre-intervention period and the period between April 2016 and May 2018 as the intervention period (Figure 1).
2.2 Intervention
Two different strategies were part of the persuasive AMS intervention: local guideline for diabetic foot infection and prospective audit and feedback on antibiotic prescriptions. The local guideline was produced by a multidisciplinary team including infectious diseases physicians, internal medicine and intensive care physicians, vascular surgeons, endocrinologists, microbiologists, pharmacists and AMS team members. The final version of the guideline was available in May 2016 and was then presented in a lecture to all physicians working in the Vascular Surgery ward. It was also released in the hospital intranet where it became available to all hospital staff. Prospective audit and feedback intervention included the review of all first prescriptions of restricted antimicrobials, all prescriptions of restricted antimicrobials lasting for more than 96 hours with the intention of changing to directed therapy and all prescriptions of antimicrobials longer than eight days. The review was performed weekly by two infectious diseases physicians that were part of the AMS team and was based on the information available in the electronic medical record. The feedback was given in a weekly face-to-face meeting with the prescribing physicians or the physician in charge of each patient. After feedback was given, a clinical discussion between AMS team and prescribing physicians ensued in order to reach a common agreed treatment strategy. The whole-hospital restrictive interventions described above were maintained unchanged during the intervention period.
The control ward, in the General Surgery department, was under the same restrictive AMS interventions described above for the whole hospital but had no defined specific AMS program.
2.3 Data collection
Ward-level data including number of admissions, type of admission (elective vs. urgent), patients’ sex and age, length of stay, 30-day readmission and in-hospital mortality were gathered through an in-house business intelligence platform described elsewhere.[16] Inpatient data on all pharmacy-dispensed antibiotics included in the Anatomical Therapeutical Chemical (ATC) group J01 were expressed as defined daily doses (DDDs)/100 patient-days according to the World Health Organization-ATC/DDD index 2013.[17] Pharmacy-dispensed antibiotics data in our hospital is corrected considering returned antibiotics. Using the antibiotics’ administration registries, the number of days with no antibiotic was obtained by subtracting the number of days with one or more antibiotic administrations from the number of hospitalization days. The indicator was expressed as a proportion of the entire length of stay. Data were aggregated by month for both wards.
2.4 Statistical analysis
Ward characteristics were described as absolute numbers (number of admissions), proportions or means (length of stay). Monthly averages for the pre- and post-intervention periods were compared for each group using T-test, Wilcoxon rank sum test and χ2 test, as appropriate. Statistical significance was assumed for p<0.05 (two tailed tests).
Carbapenem consumption, total antibiotic consumption and antibiotic free days were analysed separately for Vascular Surgery ward and for the control ward by performing segmented regression analysis of interrupted time series.[3]
We tested the change in level of each outcome and the change in slope velocity. Models’ equations are described in the supplementary file S1. We used the STATA command itsa which considers segmented linear regression. We ran the models for lag 1 after testing for autocorrelation (no significant autocorrelation was observed at higher orders). . Briefly, itsa provides the baseline value of the outcome (0), the underlying pre-intervention trend (1, using time as the predictor) and the change in the outcome level after the intervention (2, using a dummy variable defining the intervention) and the change in the slope after the intervention (3). 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.
The data were analysed with STATA version 14.0 and R software version 3.2.2.