Prospective audit and feedback on antibiotic use in neonatal intensive care: a retrospective cohort study

Our study included neonates admitted to the 55-bed inborn Level II and III NICU at Mount Sinai Hospital, started on an antibiotic and followed until discharge from the NICU. Approximately 1100 newborns are admitted annually to the unit, which cares for non-surgical preterm and sick term neonates. In 2011, prior to introducing the ASP, the rate of mortality was 5.4%; NEC was 1.9%, culture-positive early- and late-onset sepsis were 0.8 and 7.2%, respectively. Neonates with surgical conditions are sent to a tertiary pediatric hospital. Babies who did not receive an antibiotic were excluded from the study.

The pre-intervention period covered September 1, 2011, to October 21, 2012, and October 22, 2012 to November 30, 2013 reflected the ASP implementation period. Empiric therapy for early-onset neonatal sepsis (EONS, less than 72 h of age) was ampicillin and gentamicin; empiric therapy for LONS (more than 72 h of age) was cloxacillin and gentamicin, or vancomycin and gentamicin if a central catheter was present. All neonates received one blood culture when evaluated for sepsis; a second blood culture was drawn to confirm clearance if the initial blood culture was positive. Cefotaxime was added empirically if there was concern of meningitis from the clinical presentation or cerebrospinal fluid analysis.

Starting in October 2012, a pediatric infectious diseases-trained physician and/or a pharmacist specializing in antimicrobial stewardship performed PAF fives day a week on every neonate who received antibiotics. They participated in dedicated ASP rounds five mornings a week with the NICU medical team, which included staff physicians, fellows, nurse practitioners, residents and students. Neonates who were being considered for antibiotic treatment were not reviewed. The discussions included reviews of the evidence with the team, such as use of serial C-reactive protein for suspected sepsis, or clarithromycin to prevent bronchopulmonary dysplasia; discussions around microbiology and pathophysiology, such as the management of a baby born to a mother with chorioamnionitis with in-utero antibiotic exposure and negative neonatal blood cultures; support of changes in practice by individuals, such as discontinuing antibiotics with negative cultures in the context of clinical sepsis; and affirmed these changes in presentations at staff meetings. The antibiotic courses for babies followed by the infectious diseases service were not reviewed but were included in the analysis. Suggestions were made around antimicrobial use, including, but not limited to, choice of antimicrobial and duration of therapy. Eight months into the program, the frequency of ASP rounds was decreased to 4 days a week.

Antibiotic use was determined using days of therapy (DOT) and antibiotic days. DOT was calculated as the number of days the neonate was on a specific agent. For example, ampicillin for 3 days resulted in 3 DOT, irrespective of number of doses received in one day. An antibiotic day is defined as exposure to antimicrobials on a specific day (based on calendar day) and does not take into consideration the number of antimicrobials. Thus, a neonate who received ampicillin and gentamicin for 3 days would have 3 antibiotic days but 6 DOT.

Episodes were based on a count of consecutive dispense dates at the patient level and defined as suspected sepsis if antibiotic days were less than or equal to 3 to reflect empiric therapy pending culture results. If antibiotics were continued for 4 or more days, this was counted as an episode of culture-positive or culture-negative sepsis, depending on the laboratory result that accompanied the initial dispense date. If there were 3 or more days between dispense dates, then it was assumed to be a new episode.

Antimicrobial data were obtained from the Mount Sinai Hospital pharmacy database (PharmNet®, Cerner Corporation, North Kansas City, MO) for each patient, and included the name of the agent, the route of administration and the date(s) dispensed. Topical antibiotics, prophylactic antibiotics, immunizations, antifungals, antivirals and anti-retrovirals were excluded from the analyses. Patient-level pharmacy data were linked to microbiological data from the laboratory information system (SoftMic®, SCC Soft Computer, Clearwater, FL). Data on all positive cultures from sterile sites during the neonate’s admission were collected. These were also linked to clinical data extracted from the Canadian Neonatal Network (CNN) database, which uses standardized definitions to capture information from neonate medical records [25]. Study variables gathered from the database included gestational age, birthweight, mortality, diagnosis of NEC as modified Bell stage 2 or greater, and the Score for Neonatal Acute Physiology, version II (SNAP-II), a validated measure of severity of illness in a newborn during the first 24 h of admission to the NICU [18, 26, 27].

To assure data quality and adequacy, the data abstractor adhered to definitions from the CNN’s Abstractor’s Manual for clinical characteristics and outcomes of interest [25]. The database has been shown to have high reliability and accuracy [28]. Pharmacy data were repeatedly measured during and after the study period, and microbiology data were extracted by 2 independent sources to check for completeness and accuracy. Once the data were linked from the pharmacy, microbiology and CNN databases, they were manually reviewed by 2 study authors to assure data quality.

Our primary outcome measure was antibiotic DOT per 1000 patient-days (the sum of the DOT accounting for each antibiotic over 1000 patient-days). Secondary outcomes were proportion of neonates who received antibiotics during their admission; number of antibiotic days and total length of stay; mortality prior to discharge; NEC stage 2 or 3; duration of therapy for LONS, defined as an antibiotic course of 3 or more days in a symptomatic neonate aged 3 days or greater; and the incidence of antibiotic restarts within 3 days after discontinuation, as a balancing measure. We also examined duration of therapy — i.e. total consecutive antibiotic days — for culture-positive and culture-negative sepsis episodes.

We evaluated the breadth of antibiotic exposure using the Antibiotic Spectrum Index (ASI), which classifies antibiotics according to spectrum of activity, and has been shown to be sensitive to antibiotic selection for specific conditions, as compared to DOT per 1000 patient-days [29]. Using this metric, each antibiotic was scored from 1 to 13 (1 being the most narrow-spectrum agent and 13 being the most broad-spectrum agent), and a composite score of ASI per antibiotic day per patient was obtained for culture-confirmed bacteremia and culture-negative LONS.

Clinical characteristics were summarized using descriptive statistics; the Mann-Whitney U test and Pearson’s chi-squared test or Fisher’s exact test was used, as appropriate, for significance testing between the two periods.

Antibiotic DOT per 1000 patient-days pre- versus during ASP was analyzed using a Poisson model, and 95% confidence intervals were reported. This was done for the overall comparison and subgroups. Two-sided p-values less than 0.05 were considered statistically significant. All statistical analyses were performed using R statistical software version 3.4.2 (R Core Team, Vienna, Austria) [30].

The strengths of this study lie in its multi-disciplinary involvement with an infectious diseases physician and ASP pharmacist, and incorporation of PAF into the morning clinical rounds, which provided an opportunity for active engagement of neonatal staff and trainees in the discussions. Objective clinical measures were used, as well as a standardized method of reporting antibiotic consumption. Despite being a single institution study, the setting reflects many other Level III NICUs, with a broad mix of inborn non-surgical patients of variable gestational ages.

A number of limitations relate to measurement tools and the study intervention. Dispense data were acquired at the patient level, which may not have been correlated with patient administration; however, we have no reason to believe that this changed between the study periods. We did not document the content of our advice, whether recommendations were agreed upon, and associated clinical outcomes. Review of this information may have helped the ASP to target specific prescribing behaviors using other modalities in addition to PAF. There was also variability in the nature of the PAF discussions based on the readiness of each staff neonatologist to change his/her prescribing practices, which may have affected the duration of therapy and antibiotic choice at various time points.

The clinical context surrounding antibiotic courses was not available for analysis. For example, cerebrospinal fluid studies were not included, so a patient may have appropriately received cefotaxime in the context of suspected meningitis and analyzed as “culture-negative sepsis”, alongside other cases of inappropriately prolonged cefotaxime therapy. Such information would help to determine if the increase in cephalosporin use during the ASP implementation period was appropriate for the clinical context.