Does electronic stewardship work?

Abstract

Faced with national requirements to promote antimicrobial stewardship and reduce drug-resistant infections, community hospitals are challenged to make the best use of existing resources. Eighteen months after building antibiotic decision support into our electronic order platform, high-risk antibiotic use decreased by 83% (P < .001) at our community hospital. Hospital-acquired Clostridium difficile infections declined 24% (P = .07).

INTRODUCTION

Antimicrobial stewardship programs (ASPs) have been shown to reduce complications of hospital care, including drug-resistant infections, length of stay, and mortality.^1–3 ASPs have been widely integrated into care at large urban or teaching hospitals, but only in a third of community hospitals in the United States.³ One of the barriers is the considerable cost to support a robust program based on either a restricted formulary with prior authorization or prescriber audit and feedback. These core strategies are resource-intensive and require dedicated staffing.^2,3 ASPs typically include a pharmacist, an infectious disease specialist, and a microbiologist, who review trends in microbial resistance, review antibiotic orders, and track usage over time to intervene with prescribers in order to optimize antibiotic use.^2,3 With shrinking third-party reimbursements increasingly tied to performance measures, hospitals need to know where to focus improvement efforts to get results. We sought to determine whether an ASP could be successfully implemented in the community setting with limited resources. From 2012 to 2013, our hospital-acquired Clostridium difficile infection (CDI) rate was high compared to other community hospitals in our region. CDI causes severe, protracted diarrheal illness that requires treatment for weeks to months, at an average cost of $6000–15 000 per case.^4,5 Antimicrobial stewardship can significantly reduce the risk of hospital-acquired CDI.^6–9 Clindamycin and fluoroquinolones are strongly associated with CDI and increase the relative risk by 20- and 6-fold, respectively.¹⁰ We designed and implemented a prospective approval process using our hospital’s computerized order system to limit prescribing of these high-risk antibiotics.

METHODS

This project was part of a 3-year quality improvement initiative at North Shore Medical Center, a 394-bed facility in Eastern Massachusetts. Our 2 community hospitals serve different cities and the surrounding region north of Boston. Interventions were instituted sequentially and monitored over a period of 6–8 months each for impact on nosocomial CDI. In the first year, projects focused on environmental disinfection and infection-prevention measures.

In the second year, we implemented antimicrobial stewardship. Our hospitals, like many community hospitals, have no infectious-disease trained pharmacists on staff or dedicated stewardship physicians. Nor do we have software that would permit longitudinal tracking of antibiotic use. A conventional strategy of antibiotic restriction with prior authorization would have required a pharmacist to interrupt his or her workflow to communicate with clinicians each time a restricted antibiotic was ordered.¹⁶ In addition, it would have required manual data extraction and data analysis for provider audit, feedback, and reporting functions.^14,15 A prior analysis had shown that fluoroquinolones and clindamycin accounted for 25%–30% of all antibiotics dispensed at our hospital. We studied an electronic restriction on these 2 classes of high-risk antibiotics.

Restriction order sets were designed by clinicians working with nurse informaticists familiar with the order entry software. Technical analysts then built new restriction order sets into our existing computerized physician order entry (CPOE) system (2003 Siemens/Cerner Invision_tm). When a provider enters an electronic order for either clindamycin or a fluoroquinolone, a restricted antibiotic order screen appears, providing decision support. Specifying duration of treatment is a Centers for Disease Control (CDC)-recommended practice and was optional. The decision support explains the rationale for the restrictions and suggests suitable alternatives (Figure 1 ). Exceptions are permitted for approved indications. If a clinician orders the restricted drug for a nonapproved indication, a phone intervention is triggered. A review with the clinician is then initiated by a pharmacist and/or infectious disease specialist.

Figure 1.

Screen shots: antibiotic restriction order set. When a prescriber enters an order for a restricted antibiotic: (1) a restricted antibiotic order set is triggered that explains the rationale for the restrictions and suggests alternative antibiotics and (2) accepted indications for use of the restricted medication. The prescriber checks off the alternative antibiotic or the accepted indication; if the prescriber orders it for another indication, (3) he or she is prompted to enter a free-text explanation and contact the pharmacy for ongoing doses. The order goes directly to the pharmacy for review by the dispensing pharmacist.

The restriction protocol is straightforward: there are 3 approved indications for fluoroquinolones (community-acquired pneumonia, pelvic inflammatory disease/nongonococcal cervicitis/urethritis, and multidrug hypersensitivity reaction) and 5 for clindamycin (severe head/neck infection, anaerobic lung abscess, necrotizing fasciitis, toxic shock syndrome, and multidrug hypersensitivity reaction), based on professional society guidelines and the scientific literature.^11–13 If the physician orders the antibiotic for an approved indication, the medication is dispensed; if it is for a nonapproved indication, a single dose is released. The physician is then required to provide the reason for the order and contact the pharmacy for ongoing doses. If there is a discrepancy in the order, the pharmacist calls the physician for clarification. In order to accurately assess the impact of antibiotic restriction, no other changes in environmental services, hand hygiene programs, or antibiotic formulary were permitted during the study period. Physician education was provided in grand rounds presentations and at department meetings. Prior to implementation, a 3-month perioperative pilot was conducted on the surgical service. Go-live was announced by the chief medical officer in a hospital-wide communication.

Longitudinal outcomes were tracked over 3 years from May 2013 through April 2016. The primary outcome was the incidence of hospital-acquired CDIs per 1000 patient days. C. difficile outcome data for our hospital were obtained from the CDC/National Healthcare Safety Network (NHSN) reporting system.

C. difficile was defined as hospital-acquired if it appeared between the third day of admission and 30 days post discharge in a patient with no history of C. difficile within the preceding year (per NHSN criteria). Test results are obtained from our hospital microbiology laboratory, which processes all C. difficile tests for inpatients and outpatients in our service network. All test results are reported to the hospital infection control unit, which in turn reports all positive results to NHSN. For patients who are tested outside our service network, the infection control unit is contacted by the outside facility.

We measured process changes by tracking the number of computerized orders and phone interventions. Both hospital pharmacies are staffed during daytime hours and medications are dispensed directly from the pharmacy. At night, a single pharmacist covers both hospitals and a courier is dispatched between towns as needed. A limited number of antibiotics are also stocked in Omnicell_tm automated dispensers, which are present in all areas of the hospitals. Dispensed medications, including Omnicell_tm withdrawals, are tracked by the pharmacy. All pharmacy and infectious disease telephone contacts and interventions were recorded in a separate pharmacy database. Information Services (IS) generates a monthly report of all electronic orders for restricted antibiotics and the written indication, when provided.

Antibiotic usage was calculated as days on therapy (DOT) per 1000 patient days (2014 NHSN definition) based on pharmacy charge data. Alternative antibiotic use was also monitored. Prior to this initiative, our hospital had no formal ASP in place. Prescribers ordered antibiotics as clinically indicated. For the purposes of this study, we assumed that, had we started a conventional program (ie, antibiotic restriction with prior authorization), the baseline number of person-to-person interventions would have equaled the baseline number of orders for restricted antibiotics.

We tracked the total number of personnel hours required for planning and IS hours dedicated to project design, build, and testing. We continue to track the total number of hours per week required for pharmacist or infectious disease interventions. As monthly antibiotic usage reports are fully automated, ongoing IS requirements are minimal.

Statistical process control methodology was used to measure changes over time. Special cause variation was defined as events or rates 3 standard deviations above or below the baseline. Calculations of statistical significance were performed using paired t test for continuous and chi-square test for discrete variables, with a significance level of P ≤ .05. The aim of this quality improvement project was to measure the impact of a modified technical tool on aggregate clinical outcomes and did not require formal Institutional Review Board approval.

RESULTS

After go-live in November 2014, physician orders for restricted antibiotics fell by 91% (Figure 2 A), and this was sustained for 18 months. At baseline, 10 pharmacy interventions per day would have been required. After go-live, pharmacy interventions were required on average 1–2 times daily, while infectious disease consultations were requested once or twice per week

Figure 2.

Effect of electronic stewardship on hospital-acquired C. difficile infections, 2013–16. (A) A 3-month pilot of clindamycin/fluoroquinolone restrictions on the surgical service preceded go-live. Eighteen months after implementation of electronic stewardship of these high-risk antibiotics, orders fell by 91% (P < .001). (B) At 18 months, clindamycin/fluoroquinolone days on treatment per 1000 patient days fell 83% (P < .001). (C) Four months after electronic restrictions on high-risk antibiotics, hospital-acquired C. difficile infections fell 30%. In Q2 2015, increased testing of patients without clinically significant diarrhea led to an apparent increase in rates. Provider education on appropriate indications for testing led to stabilization of CDI rates at 24% below baseline (P = .07).

DOT for both classes combined decreased by 83% (Figure 2B). At 18 months, clindamycin and fluoroquinolone DOT had fallen from 15.1 to 2.4 and 48.6 to 14.4 per 1000 patient days, respectively (P < .001, P < .001). During the study period, patient length of stay remained stable: 4 days for medical and 5 days for surgical patients. The statistical process control chart (Figure 2B) demonstrates a reduction in mean DOT/1000 that was 3 standard deviations below the baseline, a change that was unlikely to be due to normal variation. This change suggests special cause variation and was temporally associated with the implementation of electronic restrictions. Alternative antibiotic use increased by 12%, evenly divided among cephalosporins, ampicillin-sulbactam, and piperacillin-tazobactam. Total antibiotic use (DOT) at our hospital did not change over time (data not shown).

At baseline, there were 90–110 hospital-acquired CDIs per year; 2 years later there were 10–20 fewer cases annually. Four months after go-live, hospital-acquired CDIs declined by 30%, from 1.16 to 0.814 cases/1000 patient days. In quarter 2, 2015, a pseudo-outbreak occurred related to increased testing with the polymerase chain reaction assay in patients without diarrhea. After provider education, rates stabilized. Eighteen months after electronic stewardship, the hospital-acquired C. difficile rate was 0.878 cases/1000 patient days, a 24% reduction from baseline (P = .07, chi-square test) (Figure 2C). After adjusting for increased polymerase chain reaction testing, our study lacked sufficient power to detect a statistically significant difference at the P = .05 level.

Startup required a total of 300 hours of combined IS, pharmacy, and infectious disease time in the planning, building, and rollout of the program. Ongoing costs were ∼5 h of person-to-person contact per week by a pharmacist or infectious disease specialist. Ongoing IS requirements relate to monthly reporting functions, which have been automated and are minimal.

DISCUSSION

Our intervention consisted of initiating a prior approval process for high-risk antibiotics through CPOE. By controlling for confounding factors over time, including other quality improvement initiatives, we were able to demonstrate a reduction in hospital-acquired CDIs associated with electronic antibiotic stewardship. The immediate decline in high-risk antibiotic orders was dramatic and was sustained for nearly 2 years. This suggests a change in physician prescribing habits, due in part to physician education, but in large part to the effect of electronic restrictions. The decision support screens continuously reinforce the education and offer alternative antibiotic choices. Additional gains are achieved through person-to-person interventions with pharmacists in real time. It is also possible that electronic stewardship exerts a deterrent effect on some prescribers. These prescribers may select a restricted antibiotic and then change their minds, knowing that they will be required to “jump through hoops” to obtain the medication. We estimate that this program has saved pharmacy resources by reducing the need for direct interventions from 10 phone calls per day to 1.5 on average. Backup interventions by the infectious disease physician are required once or twice a week.

Finally, it is important that the total clindamycin/fluoroquinolone DOT per 1000 patient days declined. This is clinically significant, as longer antibiotic treatment has been shown to increase CDIs in a dose-dependent fashion.¹⁷ And ultimately, the success of any ASP is measured in improved patient outcomes.⁸ Since 2000, the rate of hospital-acquired CDIs nationally has doubled, making it the most common health care–associated infection.^4,5 After 2 years, we were able to show a reduction in C. difficile disease, most of which was associated with electronic restrictions on high-risk antibiotics.

There have been sustained reductions in physician orders for restricted antibiotics and no apparent increase in alternative forms of medication orders (eg, paper orders). Although there was initial resistance on the part of some physicians, it largely disappeared with individual outreach. We did not perform a post-implementation survey to assess physician acceptance. However, in tracking orders by individual physicians, we saw no outliers. In addition, there was no rebound in orders for restricted medications over time. It was critical to engage physician leaders up front in planning the program as well as to ensure prescriber accountability subsequently. Whether electronic restrictions can permanently change prescribers’ appreciation of antibiotic hazards and impact future antibiotic choices is an important question that remains unanswered by the present study.

We designed a quasi-experimental study with sequential interventions. This methodology cannot control for confounding factors such as seasonal trends or intercurrent process changes that go unrecognized. Another limitation of the current study is the tracking of electronic but not paper orders. However, >95% of all medications dispensed in our hospital are ordered electronically in the CPOE system and captured in charge data. We were not able to account for orders that may have been written at the time of discharge on paper prescription pads. Chart review revealed that ∼10% of patients received a prescription for either restricted antibiotic at the time of discharge.

A third limitation is the possibility that clinicians checked off preapproved indications for nonapproved conditions. We controlled for this by having pharmacists compare orders with the available patient charts as necessary. Only the history/physical, laboratory report, and medications and allergies list are visible to pharmacists. If a discrepancy between diagnosis and treatment was noted, the pharmacist was instructed to contact the ordering physician in real time to revise the order. A second check occurred in the form of monthly reports, detailing all electronic orders and the reasons provided by clinicians. We found that, on average, work-arounds occurred 2–7 times a month. A final limitation of the present study is the lack of baseline information regarding the number of person-to-person interventions that a conventional program would have required. We assumed that personal interventions would equal the number of orders. As a result, the efficiency gained with electronic interventions may be an overestimate. Nonetheless, electronic stewardship was associated with a large and sustained reduction in the use of high-risk antibiotics.

After 18 months, a homegrown electronic ASP built on a legacy platform significantly reduced the use of high-risk antibiotics and contributed to a reduction in hospital-acquired C. difficile rates. C. difficile colitis is a severe, costly, and preventable complication of hospital care. A small up-front investment in an electronic restriction tool resulted in more efficient use of pharmacy and physician stewardship resources. Electronic stewardship has the potential to influence antibiotic selection by providers and is highly effective when combined with person-to-person interventions. It is a tool that community hospitals can design and implement locally with modest technical resources. The return on investment in terms of patient safety can be substantial.

Funding

This work was supported by the Albright-Read Institute for Health Care Improvement and Medical Research, Salem, Massachusetts, in the form of an institutional grant for quality improvement research.

Competing Interests

The authors have no conflicting/competing interests.

Contributors

Each of the authors contributed materially to the design, implementation, analysis, and interpretation of the work. Specific contributions:

Order set development/testing: BL, NK, JK, FB Jr, BS. Data management: WO'N, BL, MD. Study design and analysis: BL, NK, DP, MD, MRe, MRu. All the authors have reviewed the work critically and have given their final approval for publication. Authors acknowledge responsibility for all aspects of the work.