Approximately half of ICU patients are believed to be infected (1). These patients have an increased risk of death, and about three-quarters of them receive antimicrobials at any time (1). Antimicrobial exposure is also associated with adverse consequences, including drug-resistant and Clostridioides difficile infections. These concerns—when juxtaposed with rising antimicrobial costs and limited discovery of new antimicrobial agents—justify antimicrobial stewardship (AMS).
AMS aims to optimize antimicrobial use to improve patient outcomes while decreasing unnecessary drug use and costs (2). AMS is rapidly being adopted in healthcare facilities, facilitated by legislation and accreditation standards (3–5). Despite this, evidence-supporting AMS—especially in ICUs—is limited. The greatest benefits of hospital-based AMS programs appear to be in critical care (6), but systematic reviews of AMS programs in ICUs only identified brief, single-center studies (7 , 8), with multicenter long-term studies a major gap in evidence. A Cochrane review on AMS interventions in hospitals suggests benefit from both enabling and restrictive interventions (9). “Enabling” interventions increase the means or reduce barriers to increase capability or opportunity, whereas “restrictive” interventions use rules to reduce the opportunity to engage in the target behavior (or increase the target behavior by reducing the opportunity to engage in competing behaviors).
We conducted a phased, multisite implementation of AMS using enablement to evaluate its effects on antimicrobial consumption, costs, and patient outcomes in ICUs over a 9-year period.
MATERIALS AND METHODS
Four adult ICUs at three tertiary care general academic hospitals affiliated with the University of Toronto participated in the study. The ICUs are closed, with dedicated intensivists, pharmacists, trainees, and allied health staff. Each ICU had anywhere from four to 11 intensivists on staff over the study period, with one intensivist per team at any one time. House staff (i.e., residents and fellows who can prescribe medications) numbers were also variable, with approximately three to seven house staff per team on most days. Medical students were elective and unable to prescribe medications. Nursing ratio is 1:1 at all sites. Sites 1–3 are medical-surgical ICUs, share clinical trainees, and divide specialized services amongst them. Site 4 does not share clinical trainees with the other three ICUs. None of the ICUs previously used guidelines or other AMS tools to standardize infectious diseases (IDs) management.
Site 1 is a 16-bed ICU staffed by one team in a 472-bed hospital specializing in orthopedic and oncologic surgery, and critically ill oncology patients. Site 2 is a 24-bed ICU staffed by two teams in a 256-bed hospital specializing in neurosciences and musculoskeletal disorders. In January 2014, six neurosurgical beds were added to site 2. Site 3 is a 23-bed ICU staffed by two teams in a 471-bed hospital specializing in solid organ transplantation, respiratory, and cardiovascular diseases. Site 4 is a 20-bed cardiovascular ICU located in the same hospital as site 3 but staffed independently. The majority of patients are admitted via the emergency department. In-hospital ward transfers and out-of-hospital transfers (for expert care) comprise a varying proportion of patients, depending on the ICU.
Each of the four ICUs initiated bundles for the prevention of ventilator-associated pneumonia (VAP) and catheter-related bloodstream infections (CRBSIs) between January 2007 and March 2008, prior to the introduction of AMS (VAP and CRBSI rates were only monitored in site 1 for 3 mo prior to the introduction of AMS, precluding analysis). Infection control practices were comparable between the hospitals and did not change appreciably during the study period. No unit uses digestive or systemic decontamination.
Each ICU was served by an AMS team consisting of an ID physician and pharmacist. Six pharmacists and two physicians trained in ID participated in the intervention, although stewardship reviews were led primarily by pharmacists after the first 6 months. Physicians received stipends for the equivalent of 0.1 full-time equivalent per ICU. The AMS pharmacists were all salaried, with other AMS responsibilities outside the ICU.
The AMS interventions used enablement, focusing on feedback & monitoring and verbal persuasion (herein termed “coaching”), coupled with prospective audit and feedback of all antimicrobial prescribing (10) (for more details, see Online Supplement 1, Supplemental Digital Content 1, http://links.lww.com/CCM/E106) Feedback was provided by quarterly reporting of aggregate data for each ICU. Coaching consisted of the AMS team meeting with the ICU team to review all patients. Coaching was conducted weekdays at each site in the first year and was reduced to four times weekly the following year, and thrice weekly thereafter to reduce AMS human resource utilization. Each coaching session lasted approximately 15 minutes. Recommendations made by the AMS team were verbal. Formal ID consultation was readily available at all sites and could be suggested by the AMS team.
Introduction of AMS to Each ICU.
AMS was implemented at staggered intervals, starting with site 1 in February 2009, site 2 in December 2009, site 3 in October 2010, and site 4 in August 2012; there were no additional washout periods. Site meetings to ensure engagement and to review preintervention data took place preimplementation. In early 2012, a cluster of invasive fungal infections in heart transplant recipients in site 4 prompted routine antifungal prophylaxis in these patients.
Data Collection and Outcomes
We used retrospective data from January 1, 2007, through January 31, 2009, and prospective data from February 1, 2009, to December 31, 2015. The primary outcomes were total monthly ICU antimicrobial consumption and cost before and after introduction of AMS at each site. Antibiotics started or ordered outside of the ICU but continued upon ICU admission were included for the amount consumed in the ICU. Antibiotic consumption was measured by defined daily doses per 100 patient-days (DDDs/100), calculated monthly (11). A DDD is the average maintenance dose per day for a drug used for its main indication. For example, 2g of ceftriaxone or 14g of piperacillin (as part of piperacillin-tazobactam). Antimicrobial costs/100 patient-days were calculated based on drug-acquisition costs by the pharmacy department at each site. Drug purchasing agreements for site 1 differed from the other sites. Ceftriaxone, ciprofloxacin, and piperacillin-tazobactam experienced price reductions between 2007 and 2008, and meropenem experienced a price reduction in 2012. There were no significant drug shortages affecting ICU prescribing during the study period. Secondary outcomes include ICU mortality, length of ICU stay, ICU readmission, and standardized rates of VAP and CRBSI (12).
Escherichia coli and Pseudomonas aeruginosa drug susceptibility were assessed yearly from the microbiology information system. All antimicrobial susceptibility testing followed contemporary guidelines (Clinical and Laboratory Standards Institute [CLSI], Wayne, PA) by a shared microbiology laboratory, using first clinical isolate per patient per time period. We compared C. difficile and candidemia rates and considered first positive results in the ICU to represent ICU-onset infection.
Data on clinical factors, including demographic data, Multiple Organ Dysfunction Score, duration of mechanical ventilation, ICU length of stay, and mortality were prospectively collected on all patients as part of Ontario’s Critical Care Information System (CCIS). The CCIS is a web-based administrative database that includes information on all patients admitted to ICUs in the province of Ontario and is used for a variety of quality improvement initiatives. Site 4 did not participate in CCIS prior to July 2008.
For each ICU, patient characteristics and outcomes before and after the intervention were summarized and 95% CIs were calculated for differences between the two time periods. Antimicrobial consumption and cost were based on unit-based monthly summaries. The data are structured as monthly time series with repeated measures of utilization for each drug. The data were hierarchically clustered, with each utilization measure derived from one of four sites. We used the methods of Hussey and Hughes (13) and applied linear mixed effects models. Introduction of the AMS was the sole fixed effect, and we estimated a random slope and a random intercept for each drug nested within each ICU. The adjusted intervention effect is the average expected change in utilization at the onset of the intervention, accounting for site level heterogeneity in level of utilization (random intercepts) and site-specific temporal trends (random slopes). We used linear effects modeling for additional sensitivity analysis and reanalyzed the data using only prospectively collected data (i.e., after February 2009), using marginal effects (conditional on random slope and intercept for each site).
We examined antibacterial and antifungal consumption as well as cost, and grouped drugs into antimethicillin-resistant Staphylococcus aureus (MRSA) and antipseudomonal agents to help demonstrate changes accounting for the effect of the AMS intervention.
For secondary outcomes, we used generalized linear models (one for each combination of site and outcome), adjusting for time. Each model includes one slope and a jump discontinuity at the start of the intervention. We used unadjusted Poisson tests for candidemia and C. difficile rates before and after the intervention.
There are no accepted methods for measuring clinically relevant changes in antimicrobial resistance over time. As such, we chose to use heat maps (i.e., color-coded matrices of resistance over time, where green and red reflect more and less susceptible, respectively) to display P. aeruginosa and E. coli susceptibility by year for commonly used antimicrobials (14). In September 2010, there was a recall of susceptibility testing cards for piperacillin-tazobactam for the VITEK 2 system (bioMérieux, MArcy-l'Étoile, France) due to quality issues dating back to December 2009. In 2012, CLSI changed susceptibility breakpoints for piperacillin-tazobactam and P. aeruginosa. These events prevent appropriate comparisons of piperacillin-tazobactam susceptibility for the study period. Additionally, the laboratory changed the method of testing for C. difficile from an enzyme immunoassay (Premier Toxins A&B; Meridian Biosciences, Cincinnati, OH) to a more sensitive polymerase chain reaction-based assay (Xpert C. difficile/Epi; Cepheid, Sunnyvale, CA) in January 2011.
The Research Ethics Boards of each institution approved the study.
Over the 9-year study period, a total of 239,123 patient-days (57,195 patients) were analyzed: 42,142 patient-days at site 1, 76,148 patient-days at site 2, 66,888 patient-days at site 3, and 53,942 patient-days at site 4. A total of 148 832 patient-days occurred after the introduction of the AMS: 32,202 patient-days at site 1, 54,824 patient-days at site 2, 40,582 patient-days at site 3, and 21,224 patient-days at site 4.(Table 1) Patient age, gender, and severity of illness were comparable before and after introduction of AMS with few exceptions.
Following introduction of AMS in each of the ICUs, total antibacterial use decreased 12% (120.90 vs 110.50 DDDs/100 patient-days; 95% CI, –16.75 to –7.49; p < 0.001) (Table 2; and Online Supplement 2, Supplemental Digital Content 2, http://links.lww.com/CCM/E107) Among these, there was decreased use of anti-MRSA drugs (12.77 vs 10.52 DDDs/100 patient-days; 95% CI, –3.47 to –1.04; p < 0.001) and antipseudomonal drugs (43.72 vs 34.05 DDDs/100 patient-days; 95% CI, –12.78 to –6.56; p < 0.001). Total antifungal use decreased 4% (30.53 vs 27.37 DDDs/100 patient-days; 95% CI, –8.33 to 0.04; p = 0.05).
Total antibacterial cost in the ICUs decreased 20% ($3195.56 vs $1998.59/100 patient-days; 95% CI, –$905.85 to –$378.84; p < 0.001).(Table 2; and Online Supplement 2, Supplemental Digital Content 2, http://links.lww.com/CCM/E107) There were decreased costs of anti-MRSA drugs ($323.36 vs $261.76/100 patient-days; 95% CI, –$108.97 to –$14.23; p = 0.01) and antipseudomonal drugs ($530.92 vs $151.08/100 patient-days; 95% CI, –$592.38 to –$167.30; p < 0.001). Total antifungal costs did not significantly decrease ($558.44 vs $324.79/100 patient-days; 95% CI, –$661.69 to $194.39; p = 0.28).
Results of the sensitivity analysis using only prospective data show similar direction and magnitude of changes in antibacterial and antifungal consumption and cost in sites 2 to 4 (Online Supplement 3, Supplemental Digital Content 3, http://links.lww.com/CCM/E108).
There were no differences in ICU mortality at any of the sites following introduction of AMS (Table 3). Rates of VAP were 32% lower post intervention (4.56 vs 1.61 cases/1,000 mechanical ventilation days; 95% CI, –2.84 to –0.15; p = 0.01). Other secondary outcomes are shown in Table 3.
Antimicrobial Resistant Organisms
There were no consistent effects on P. aeruginosa and E. coli susceptibility from the intervention (Supplemental Fig. 1A, Supplemental Digital Content 4, http://links.lww.com/CCM/E109; legend, Supplemental Digital Content 6, http://links.lww.com/CCM/E111). As mentioned, assessment of piperacillin-tazobactam susceptibility was not possible. Similarly, inconsistent susceptibility patterns were observed for E. coli (Supplemental Fig. 1B, Supplemental Digital Content 4, http://links.lww.com/CCM/E109; legend, Supplemental Digital Content 6, http://links.lww.com/CCM/E111). Candidemia rates were unchanged at all of the sites except site 1 (3.16 vs 1.89 cases/1,000 patient-days; p = 0.04) (Online Supplement 4, Supplemental Digital Content 5, http://links.lww.com/CCM/E110). Changes in nosocomial C. difficile rates were uninterpretable because the method of testing changed during the study period.
We used data from four academic hospital ICUs covering 9 years and almost 240,000 critical care patients-days. We found that enablement-based AMS is associated with reduced antimicrobial consumption and costs, without worsening mortality, and with no appreciable change in antimicrobial resistance or candidemia. The reduced antimicrobial consumption and costs are associated with reduced use of antipseudomonal, anti-MRSA, and antifungal agents following AMS introduction.
The strengths of this study include multiple sites, a large number of patients, the varied ICUs, long-term observation following implementation of a quality improvement initiative minimally dependent on technology, and the robust nature of the results. This study is the largest and longest such undertaking involving AMS in ICUs. It is also the first to convincingly demonstrate reductions in utilization and cost without “squeezing the balloon,” where targeting use of one drug or drug class is counterbalanced by an increase in use of another drug or class (15). We demonstrate that the intervention can be effective in a wide range of intensive care settings. Using mortality, length of stay, and readmission as safety signals, we also show that this intervention is neither associated with patient harm nor obvious clinical benefit.
The study has limitations. First, it is an observational study without randomization and thus is subject to known biases, especially secular trends (i.e., changes occurring over a long term, such as institutional learning). We sought to overcome this with a phased implementation resulting in numerous data points before and after the intervention to help strengthen the experimental association. The observed reduction in reported VAP could be causally related to reduced antimicrobial use; however, such a relationship has not been shown previously. Also, because VAP and CRBSI are self-reported by each ICU into an administrative database, we could not independently adjudicate the quality of this reporting. The converse is also true: an effort to reduce VAP may have reduced the need for antimicrobials. However, the timing of introducing VAP interventions (January 2007 to March 2008) and the timing of the observed reductions in antimicrobials fits an effect of AMS on antimicrobials rather than VAP prevention as the mechanism for reduced antimicrobial consumption. Second, this study occurred in academic ICUs, which differ from community-based ICUs in terms of staffing models and having more complex patients receiving broader spectrum antimicrobials. Third, most of the cohort data were prospectively collected, but a small amount was retrospectively assembled. Comparing these data separately in the model could not be performed because of the number of observational requirements for time-series analyses. Fourth, although this study used important safety indicators to identify potential harm of the intervention, it is possible that we may not have measured important causes of harm, such as IDs-specific morbidity and mortality that may not be apparent when measuring overall ICU outcomes. We think this is unlikely, however, with studies being consistent that AMS is not associated with harm (9). This study represents the efforts of an established AMS program with appropriate resourcing for ICU-based coaching. Although many healthcare providers were involved in the intervention, the results may not be generalizable to other hospitals and AMS programs. Microbiology diagnostic testing issues also limit the interpretation of some of our results. The mid-study change to a more sensitive assay for C. difficile precluded informative analysis (16). Additionally, quality issues with piperacillin-tazobactam susceptibility testing were recognized by the manufacturer over a year after problems emerged and, along with changes in thresholds, prevent making firm conclusions around changes in susceptibility.
In one of the largest and highest quality studies of AMS in ICUs, Elligsen et al (17) evaluated the effect of audit and feedback for broad-spectrum antibiotics in three ICUs in a single center, with 12 months of observation preceding and following AMS introduction, involving roughly 4,700 patients. Using interrupted time-series analysis, they showed an approximately 22% reduction in broad-spectrum antibiotic prescribing, a 13% reduction in overall antibiotic prescribing, and a 24% reduction in costs. Antimicrobial resistance improvement was shown only to meropenem. Similarly, Karanika et al (6) found that AMS could reduce consumption in ICUs by 39% and costs by 34%. Our study—which used enablement, and addressed all antimicrobials rather than only broad-spectrum antibacterial agents—shows comparable reductions in antibacterial consumption to Elligsen et al (17), with additional overall antimicrobial usage benefits.
Potential effects of our intervention may be underestimated with our analysis. Indeed, antibacterial consumption was significantly decreased in unit 3, but there was no significantly decreased cost. The crude reduction in cost was $2,242/100 patient-days, but this became insignificant when adjusted, with CI crossing 0. Additionally, because of differences in drug-acquisition costs, drug-resistance, and healthcare systems structure and funding, results may not be generalizable outside of Canada.
A systematic review of AMS in ICUs from 1996 to 2010 (7) found that AMS likely improves prescribing with accompanying beneficial effects on ICU resistance rates. However, important gaps in study quality were identified. Mertz et al (8) reviewed AMS in ICUs from 2010 to 2015, and found ongoing poor study quality, emphasizing limited duration of observation, and lack of patient-specific outcomes and antimicrobial resistance effects. A recent systematic review and meta-analysis of all hospital-based AMS programs from clinical and economic perspectives—six of which were in ICUs (6)—found that most studies were brief, from single centers. Our study specifically aims to fill in these gaps by evaluating the effects of AMS in ICUs in multiple hospitals, with follow-up ranging from 3.5 to 6.8 years. Importantly, we demonstrate clinically relevant microbiologic stability over a prolonged period of sustained AMS, consistent with our preliminary findings (18). Although AMS is primarily aimed at reducing antimicrobial resistance, we only looked at specific patterns of drug-resistance. Demonstrating overall reductions in acquisition of drug-resistance should be a priority for future work.
Behavioral change techniques in AMS in the United States are primarily performed by pharmacists, although there is a widespread belief that physician intervention substantially improves AMS success (19). Audit and feedback and academic detailing have been previously shown to improve physician performance and antimicrobial prescribing (9 , 20). This study shows that persuasive techniques—primarily coaching by AMS pharmacists—can sustainably improve antimicrobial prescribing in ICUs.
Antimicrobial prescribing is primarily inappropriate due to initiation, spectrum, or duration. Most studies of AMS prospective audit and feedback focus on “restricted” or “targeted” drugs—and so only tackle the problem of inappropriately broad-spectrum therapy. Our intervention aimed to improve all antimicrobial prescribing. As such, unnecessarily prolonged therapy would be audited and presented to the prescriber, as would use of a too narrow-spectrum agent. Reductions in overall antimicrobial use, as well as reductions in specific antimicrobial drug classes, suggest that this approach worked. The fidelity of AMS was not consistent across all ICUs (although all ICUs saw crude reductions in antibacterial consumption). There are many potential reasons for this, including the possibility that AMS using behavior change techniques is not a universally beneficial intervention. We suspect that low baseline antimicrobial use (as in ICUs two [neuro focused unit] and four [cardiovascular focused unit]) can reduce the potential impact of AMS. Because of high patient turnover, reductions in antimicrobial use may not be possible when length of stay is short, as in ICU four. We consider all antimicrobial prescribing important and have shown that it is feasible to work with intensivists to improve antimicrobial prescribing by considering all patients, and their accompanying microbiology laboratory results and antimicrobial prescriptions (21). Additionally, because we maintained a minimum intervention frequency of thrice weekly, we were able to frequently intervene around the time of prescribing and at regular intervals.
Despite the need for AMS in acute care institutions, inadequate funding of AMS personnel is an important barrier to stewardship implementation (19 , 22). Our intervention requires dedicated personnel. Estimating costs of the ICU-based program is difficult because AMS programs (like infection prevention and control programs) are hospital-wide, and thus do not only support ICUs. Although we regularly demonstrate to our hospital leaders that hospital-wide AMS savings outweigh AMS costs, we are unable to refine the analysis to ICU-specific work. Our work, however, does demonstrate that there are potential cost-savings associated with AMS that can help offset the cost of program development and support.
AMS in ICUs with coaching plus audit and feedback is associated with sustained improvements in antimicrobial consumption and cost. ICUs with high antimicrobial consumption or expenditure should consider implementing AMS programs. The importance of AMS is primarily improving patient safety by getting the right antibiotic to the right patient when they need it, and only when they need it.
We are indebted to Patrick Cheng, University Health Network, and Melanie Thomson, Sinai Health System for their help in collating the data for this study.
1. Vincent JL, Rello J, Marshall J, et al; EPIC II Group of Investigators: International study of the prevalence and outcomes of infection in intensive care units. JAMA 2009; 302:2323–2329
2. Barlam TF, Cosgrove SE, Abbo LM, et al. Implementing an antibiotic stewardship program: Guidelines by the Infectious Diseases Society of America and the Society for Healthcare Epidemiology of America. Clin Infect Dis 2016; 62:e51–e77
3. Goff DA, Kullar R, Bauer KA, et al. Eight habits of highly effective antimicrobial stewardship
programs to meet the Joint Commission standards for hospitals. Clin Infect Dis 2017; 64:1134–1139
4. Accreditation Canada: Required Organizational Practices Handbook 2017. 2016Ottawa, ON, Canada, Accreditation Canada.
5. Joint Commission on Hospital Accreditation: APPROVED: New antimicrobial stewardship
standard. Jt Comm Perspect 2016; 36:1, 3–4, 8
6. Karanika S, Paudel S, Grigoras C, et al. Systematic review and meta-analysis of clinical and economic outcomes from the implementation of hospital-based antimicrobial stewardship
programs. Antimicrob Agents Chemother 2016; 60:4840–4852
7. Kaki R, Elligsen M, Walker S, et al. Impact of antimicrobial stewardship
in critical care
: A systematic review. J Antimicrob Chemother 2011; 66:1223–1230
8. Mertz D, Brooks A, Irfan N, et al. Antimicrobial stewardship
in the intensive care setting–a review and critical appraisal of the literature. Swiss Med Wkly 2015; 145:w14220
9. Davey P, Marwick CA, Scott CL, et al. Interventions to improve antibiotic prescribing practices for hospital inpatients. Cochrane Database Syst Rev 2017; 2:CD003543
10. Michie S, Richardson M, Johnston M, et al. The behavior change technique taxonomy (v1) of 93 hierarchically clustered techniques: Building an international consensus for the reporting of behavior change interventions. Ann Behav Med 2013; 46:81–95
11. Morris AM, Brener S, Dresser L, et al. Use of a structured panel process to define quality metrics for antimicrobial stewardship
programs. Infect Control Hosp Epidemiol 2012; 33:500–506
12. Ventilator associated pneumonia and central line infection prevention toolkit. 2012Toronto, ON, Canada, Critical Care
13. Hussey MA, Hughes JP. Design and analysis of stepped wedge cluster randomized trials. Contemp Clin Trials 2007; 28:182–191
14. Hughes JS, Hurford A, Finley RL, et al. How to measure the impacts of antibiotic resistance and antibiotic development on empiric therapy: New composite indices. BMJ Open 2016; 6:e012040
15. Burke JP. Antibiotic resistance–squeezing the balloon? JAMA 1998; 280:1270–1271
16. Hernández-Rocha C, Barra-Carrasco J, Álvarez-Lobos M, et al. Prospective comparison of a commercial multiplex real-time polymerase chain reaction and an enzyme immunoassay with toxigenic culture in the diagnosis of Clostridium difficile-associated infections. Diagn Microbiol Infect Dis 2013; 75:361–365
17. Elligsen M, Walker SA, Pinto R, et al. Audit and feedback to reduce broad-spectrum antibiotic use among intensive care unit patients: A controlled interrupted time series analysis. Infect Control Hosp Epidemiol 2012; 33:354–361
18. Hurford A, Morris AM, Fisman DN, et al. Linking antimicrobial prescribing to antimicrobial resistance
in the ICU: Before and after an antimicrobial stewardship
program. Epidemics 2012; 4:203–210
19. Livorsi DJ, Heintz B, Jacob JT, et al. Audit and feedback processes among antimicrobial stewardship
programs: A survey of the Society for Healthcare Epidemiology of America Research Network. Infect Control Hosp Epidemiol 2016; 37:704–706
20. Ivers N, Jamtvedt G, Flottorp S, et al. Audit and feedback: Effects on professional practice and healthcare outcomes. Cochrane Database Syst Rev 2012; 6:CD000259
21. Holmes AH, Moore LS, Sundsfjord A, et al. Understanding the mechanisms and drivers of antimicrobial resistance
. Lancet 2016; 387:176–187
22. Pollack LA, van Santen KL, Weiner LM, et al. Antibiotic stewardship programs in U.S. acute care hospitals: Findings from the 2014 National Healthcare Safety Network Annual Hospital Survey. Clin Infect Dis 2016; 63:443–449