Empiric Vancomycin Reduction in a Pediatric Intensive Care Unit

Nationwide Children’s Hospital is a 527-bed, tertiary care, freestanding academic hospital in Columbus, Ohio, with a 54-bed multidisciplinary, quaternary PICU with >3000 annual admissions. The local prevalence of MRSA during our study period from patients throughout the hospital was 48% and 46% in 2017 and 2018, respectively. Patients who received empirical antibiotic therapy in the PICU for suspected CAIs were eligible for the QI study. QI data specialists extracted patient data from the electronic medical record (EMR). To identify patients with suspected CAIs, we first identified all patients who received an intravenous antibiotic (excluding prophylaxis) within the first 48 hours after admission to the PICU from the emergency department or an inpatient unit. We then excluded patients with risk factors for nosocomial or health care-associated infections, including (1) those who had been hospitalized for >48 hours before their suspected infection, (2) those who had recent hospitalization (within the previous 30 days) or who reside in a chronic care facility, and (3) those with chronic comorbidities (cystic fibrosis, active malignancy or organ transplant receipt, tracheostomy dependent or chronic ventilator requirement, immunosuppressed). In addition, we excluded patients who were prescribed antibiotics included in our PICU’s protocol for suspected antibiotic-resistant infections, which includes all the aforementioned risk factors as well as patients with recent antibiotic use (within last 6 weeks) or history of resistant organisms. We also excluded patients transferred to our PICU from outside institutions because outside hospital records were not available to our QI team. After exclusions, we presumed the remaining patients received intravenous antibiotic therapy for suspected bacterial CAIs. We performed an initial validation of the extracted data by manual chart review.

A multidisciplinary QI team formed in March 2018, including pediatric infectious diseases fellows, a pediatric critical care physician and pharmacist, a pediatric infectious diseases physician and pharmacist, and QI specialists. We developed consensus indications for empirical vancomycin use for PICU patients with suspected bacterial CAIs based on expert opinion, reported literature, and local S aureus infections in the PICU. During their development, we reviewed 3 years of S aureus infections (32 of 128 [25%] were MRSA) among patients with suspected CAIs admitted to our institution’s PICU to ensure our indications were applicable to our patient population and local epidemiology of S aureus infections. The indications addressed clinical and epidemiological risk factors for infection with either S aureus or ceftriaxone-resistant Streptococcus pneumoniae (ie, meningitis) and included the following: (1) septic shock; (2) central nervous system infection; (3) complicated or extensive pneumonia; (4) skin, soft tissue, deep neck, or musculoskeletal infection; (4) personal or family history of MRSA; and (5) confirmed concurrent influenza infection.^17,22,23  We then retrospectively applied these indications to 91 PICU patients with suspected CAIs who received vancomycin (or linezolid) from May 13, 2017, to September 27, 2017. Of note, these 91 patients had been previously identified via query of the EMR, before the more comprehensive data extraction from our QI data specialists, and thus represent an incomplete subset of the patients ultimately included in the background data. Application of our vancomycin indications to the 91 patients demonstrated a potential to reduce empirical vancomycin use by a relative 43% without missing any MRSA infections. Therefore, in May 2018, our QI team developed a key driver diagram with the aim to decrease empirical vancomycin use among PICU patients with suspected bacterial CAIs from a baseline of 73% (May 2017 to April 2018) to 51% (a relative 30% reduction) by December 2018 (Fig 1).

In May 2018, we presented the study background, rationale, and aim to the division of critical care medicine, who as a group concurred with the consensus vancomycin indications and agreed to apply them to patient care. As a visual reminder for clinicians, paper placards (Supplemental Fig 5) listing empirical vancomycin indications were posted on workstations in the PICU. In June 2018, we began educating medical residents about the QI study during their orientation to the PICU 4-week rotation. In July 2018, we modified a preexisting PICU empirical antibiotic order set in the EMR with passive decision support listing the indications for empirical vancomycin.

Our primary outcome metric was the percentage of patients with suspected bacterial CAIs who received vancomycin within 48 hours of admission to the PICU, tracked retrospectively from May 2017 through April 2018 (baseline) and prospectively from May 2018 through July 2019. Overall vancomycin days of therapy per 1000 patient days in all PICU patients was a secondary measure. We also tracked the development of AKI; patients were considered to have kidney injury if they had an increase in serum creatinine from their initial value of at least 50% within 7 days after PICU admission.⁹  Baseline creatinine was considered as the first value for that medical encounter. We also tracked 2 balancing measures: (1) use of clindamycin or linezolid in the first 48 hours of PICU admission (as a replacement for vancomycin) and (2) positive culture results or polymerase chain reaction (PCR) results for S aureus, ceftriaxone-resistant S pneumoniae, or Enterococcus spp. in patients who did not receive empirical vancomycin.

Data were extracted from the EMR and tracked on Shewhart statistical process control charts by using a locally developed plug-in for Microsoft Excel. The baseline period was defined by the 12-month period before the first intervention. We then applied rules from the American Society for Quality (https://asq.org/quality-resources/control-chart#Procedure) to identify special cause variation and define sequential process stages. In each process stage, the mean was calculated by using the data points that defined criteria for a centerline shift (eg, 8 in a row below the previous centerline; 10 of 11 below the previous centerline). Calculation of control limits was based on process average and the subgroup size, with limits based on a binomial distribution of the data. As a QI initiative, this study did not meet the definition of human subjects research and therefore did not require institutional review board review or consent. We followed the Standards for Quality Improvement Reporting Excellence 2.0 guidelines in reporting this QI study.²⁴  Creatinine data were analyzed by using Fisher’s exact test.

From May 2017 through July 2019, we identified 1620 PICU patients with suspected bacterial CAIs, ranging from 34 to 94 patients each month. The percentage of patients who received empirical vancomycin decreased from a baseline mean of 73% (524 of 722, May 2017 to April 2018) to a new process stage mean of 45% (187 of 420, May 2018 to December 2018) during the postintervention period, a 38% relative reduction (Fig 2). During the same time period, overall vancomycin use in the PICU (ie, vancomycin use for any indication, not limited to suspected CAIs) decreased from an average of 343 to 218 days of therapy per 1000 patient days (Fig 3). Of the 1620 patients included in our study, 883 were part of the postintervention period. Of these, only 880 had >1 creatinine value documented. There was no significant difference in the proportion of patients with a 50% or greater increase in creatinine concentration during the baseline versus postintervention periods (105 [14.2%] of 737 patients versus 137 [15.6%] of 880 patients, P = .48, Fischer’s exact test).

The use of empirical clindamycin or linezolid in the first 48 hours decreased from a baseline mean of 7.2% to 4.9% during the postintervention period (Fig 4). Of 883 patients with suspected CAIs in the postintervention period, 108 (12.2%) had a positive culture result or PCR result for S aureus, S pneumoniae (all susceptible to ceftriaxone), or Enterococcus spp (Table 1). Of those, only 63% received empirical vancomycin. Of the 40 patients who did not receive empirical vancomycin, 26 received appropriate treatment with an alternative antibiotic, and 14 were not treated for the detected organism because it was thought to represent colonization by the treating clinician. There were no infections for which vancomycin was required afterward but not received empirically.

In our study, we used QI methodology to complement an existing antimicrobial stewardship program that routinely performs prospective audit and feedback for vancomycin use. Prospective audit and feedback is a successful tool recommended by national guidelines for antimicrobial stewardship programs to improve antimicrobial use.^25–27  However, this requires dedicated pharmacists and physicians for those active interventions. Also, prospective audit and feedback addresses vancomycin that has already been ordered and, in most instances, administered to the patient. Even approaches that restrict vancomycin at the ordering stage failed to reduce inappropriate empirical vancomycin use in one multicenter study.²⁸  In this QI initiative, we identified an opportunity to reduce vancomycin use in the PICU by preventing it from being ordered empirically. With a QI approach, we complemented the antimicrobial stewardship program’s efforts and reduced substantially the percentage of patients with CAIs who received empirical vancomycin. We chose a more conservative goal that the identified potential for reduction; however, as the results show, we exceeded that goal. Although we cannot directly prove causality, this was the only new intervention targeting vancomycin use in our PICU during that time period. Concurrently, we saw an overall decrease in vancomycin days of therapy in the PICU despite no other new interventions or changes to standard practice.

Developing clear indications for empirical vancomycin use made the project feasible and easy to apply by all staff members, because nearly all information for risk assessment was already routinely available. Our risk factors, although not yet validated in children, appear to perform well in identifying those children for whom vancomycin can be excluded from their empirical antibiotics. Similar approaches have had success. Kim et al reduced empirical vancomycin use in neutropenic fever through an education effort, and Chiu et al reduced vancomycin use in the neonatal ICU by developing guidelines for vancomycin use.^29,30  Although those authors used national guidelines as reference to change their practice, in the absence of a national guideline for empirical vancomycin use in the PICU, local validation was crucial to our success. Our empirical vancomycin indications were validated locally by reviewing our institution’s PICU patients and S aureus epidemiology, which gave our indications reliable applicability in our patient population. Changing practice habits is always a potential barrier for stewardship interventions, especially in the setting of critically ill patients. However, our data-driven approach enabled us to confirm that our indications were applicable to our specific patient population and, most of all, that they were safe. Demonstrating to the PICU physicians with local data the safety of our new proposed indications was essential for their buy in and the success of the project.

Balancing measures indicated that our interventions did not result in unintended consequences or inadequate treatment of infections. With decreased use of vancomycin, we did not observe a compensatory increase in the use of empirical linezolid or clindamycin as alternative agents for MRSA coverage. In fact, use of both antibiotics decreased. This, in conjunction with the reduction in vancomycin use, suggests that there was widespread clinical acceptance of the guideline among our PICU clinicians. We also did not observe any clinically relevant microbiologic results suggesting scenarios in which vancomycin was necessary but not given.

Despite our successful decrease in vancomycin use, we did not see a significant change in the prevalence of AKI in the PICU. This was similar to the findings of Hsu et al³¹  in which they also did not see a significant change in the prevalence of acute renal insufficiency despite initiatives to reduce vancomycin use. This likely represents a multitude of factors that contribute to acute renal insufficiency, such as duration of vancomycin exposure, presence of comorbidities, severity of illness, sepsis, and other nephrotoxic drugs.^31,32  It is also possible that our limited definition of AKI underestimated the true AKI rates in our population.

Our QI study does have limitations. First, our indications have not yet been validated. We are currently planning formal validation in a larger cohort of critically ill children. We also focused solely on empirical vancomycin use, which is only one aspect of appropriate vancomycin use. Additional opportunities for QI may include de-escalation or discontinuation based on microbiologic test results, as well as duration of therapy for confirmed infections, and using AUC levels for drug monitoring.^25,33–35  Second, we excluded patients with known risk factors for health care–associated infections, and our interventions would not be appropriate to apply to that group of patients without modification. Third, our balancing measure to evaluate replacement of vancomycin with other MRSA-active antibiotics was limited to linezolid and clindamycin. We did not track other potential antibiotics such as ceftaroline or daptomycin, but their use is tracked by the antimicrobial stewardship program and is infrequent.²³  Lastly, this was a single-center study based in an area with high prevalence of MRSA. These indications may not be reproducible in areas with lower prevalence of CA-MRSA and could potentially lead to overuse of empirical vancomycin. Further next steps are reviewing in more depth the patients who receive vancomycin to assess its appropriateness and if they met our proposed indications. This will provide valuable insight in how we can further improve our vancomycin use.

A multidisciplinary QI approach with consensus empirical vancomycin indications successfully and safely reduced vancomycin use in our large academic center PICU. Validating the consensus indications in our own, local population and obtaining buy-in from front-line, critical care providers were essential to successful implementation of the indications and subsequent decreased vancomycin use.

AKI

acute kidney injury

CAI

community-acquired infection

CA-MRSA

community-acquired methicillin-resistant Staphylococcus aureus

EMR

electronic medical record

MRSA

methicillin-resistant Staphylococcus aureus

PCR

polymerase chain reaction

QI

quality improvement