National Healthcare Safety Network 2018 Baseline Neonatal Standardized Antimicrobial Administration Ratios Free

Patient-level data are not reported to the AU Option, and only location- and hospital-level data are available for risk adjustment of AU rate data. Because neonatal AU risk factors differ from adult and pediatric risk factors, we added questions to the NHSN Patient Safety Component Annual Hospital Survey¹⁰  to capture hospital-level data needed for risk adjustment. The ASP-SIG assessed which neonatal risk factors were most important for risk adjustment and were available to all hospitals for electronic capture (Supplemental Table 4). After identifying a list of factors, the group developed, tested, and fine-tuned neonatal survey questions. Neonatal questions were first added to the 2018 NHSN Annual Hospital Survey, which hospitals completed in early 2019. Survey questions asked hospitals to report annual admissions (specifically, inborn versus outborn and by birth weight [BW] categories) and data on the level of neonatal care provided.¹⁰ 

During monthly meetings, the ASP-SIG identified individual and groups of antimicrobial agents that represented important stewardship targets; these became the 2018 baseline neonatal SAAR antimicrobial agent categories. Categories include (1) vancomycin predominantly used for treatment of late-onset sepsis, (2) broad-spectrum antibacterial agents predominantly used for hospital-onset infections, (3) third-generation cephalosporins, (4) ampicillin predominantly used for treatment of early-onset sepsis, (5) aminoglycosides predominantly used for the treatment of early-onset and late-onset sepsis, (6) fluconazole predominantly used for candidiasis, and (7) all antibacterial agents. Indication for AU is not reported to the AU Option; therefore, we cannot say for certain whether antimicrobial agents in each category were used for treatment, prophylaxis, or some other reason. Details on specific antimicrobial groupings can be found in Appendix E of the NHSN AUR Module Protocol.⁷ 

In 2019, when neonatal SAARs were developed, hospitals could report to the AU Option from 4 NHSN-defined¹¹  neonatal patient care locations: (1) level I well newborn nurseries, (2) level II special care nurseries, (3) level II/III NICUs, and (4) level III (and/or IV) NICUs. The NHSN-defined level II/III NICU is a combined nursery that houses both level II and III newborns and infants, analogous to a mixed acuity unit specifically for neonatal critical care patients. Definitions of other units can be found in the NHSN location mapping document.¹¹  We analyzed rates of AU in each of these location types for all SAAR antimicrobial agent categories and found that rates in level I units were too low for inclusion in SAAR models and that level II fluconazole rates were too low for inclusion in the fluconazole SAAR model. The 2018 baseline neonatal SAAR referent population included level II, level II/III, and level III (and/or IV) neonatal units reporting at least 9 months of data in 2018. In December 2019, NHSN created 2 distinct locations types for level III and level IV NICUs, allowing hospitals to report data for each separately. Before this distinction, hospitals included level IV data in with level III data when reporting to NHSN. SAARs are available for both level III and IV units; these units have identical risk adjustment because they were modeled together as one.

NHSN AU data were reviewed and validated using the same process described previously,⁹  and records with errors in the AU Option numerator (antimicrobial days of therapy)⁷  or denominator (days present)⁷  were excluded from predictive modeling. The 2018 NHSN Annual Hospital Survey was used to characterize hospitals, and data were reviewed for potential errors. Hospitals that reported neonatal AU data but responded “my facility does not provide neonatal patient care services” on the survey were excluded from analyses, as they were not required to complete the neonatal section of the survey, which was necessary for risk adjustment. Hospitals reporting level II or higher AU data but reporting 0 level II or higher annual neonatal admissions on the survey were excluded from further analyses. Finally, hospitals that reported improbable annual admissions by BW category were excluded from further analyses. For example, if a hospital were to report all 2018 neonatal admissions as ≤750 g but only include level II units in its NHSN reporting plan, the hospital would be excluded.

We pooled antimicrobial days of therapy across agents within each SAAR agent category for each location/month. Next, we summed antimicrobial days and days present across all months, so analysis datasets contained 1 record for each location and SAAR agent category (location/year level). We used forward stagewise negative binomial regression to assess associations between AU rates and potential risk factors for each SAAR antimicrobial agent category.

Candidate risk factors included location type, location type plus other levels of neonatal care reported to the AU Option (variable called “other levels”), facility type, medical school affiliation, hospital teaching status, total number of hospital beds, whether the hospital provided level III or higher neonatal care, whether the hospital accepted neonates as transfers for complex procedures specified in the survey, number of annual neonatal admissions, number of annual inborn admissions, number of annual outborn admissions, percentage of annual admissions that were outborn, number of annual admissions of infants weighing ≤750 g, 751–1000 g, 1001–1500 g, 1501–2500 g, and >2500 g, and percentage of annual admissions that were normal BW (>2500 g), low BW (1501–2500 g), or very low BW (VLBW) (<1501 g). For the other levels variable, we assessed location type plus information on what other levels of neonatal care were reported to the AU Option. We created this variable to assess whether AU in a level II unit, for example, differs between hospitals that only provide level II care and those that additionally provide level II/III, III, or IV care.

All continuous variables were assessed as deciles, quintiles, quartiles, tertiles, and at the median. Covariates were assessed for multicollinearity. As with 2017 baseline SAAR model development, 2 analysts worked independently to develop each of the 7 neonatal SAAR predictive models to maximize objectivity during model development. Additional detail on the dual analyst process can be found in the methods section of our article on the 2017 baseline SAAR.⁹  All analyses were conducted using SAS version 9.4 software (SAS Institute, Inc, Cary, NC). SAAR models were evaluated using Akaike and Bayesian information criteria and likelihood ratio χ² tests. Models that resulted in strata with only 1 location were reconsidered to prevent imprecision in model estimates. Final models were tested for influential observations and validated using bootstrap resampling methods.

The SAAR neonatal referent population, after exclusions, included 324 units reporting from 304 hospitals (56 level II nurseries, 152 level II/III NICUs, and 116 level III [and/or IV] NICUs). Units were reported from children’s, general acute care, military, women’s, and women’s and children’s hospitals. The median number of hospital beds was 319, and ∼50% were major teaching hospitals. Hospitals had a median of 316 neonatal admissions per year, the majority of which were inborn (median outborn, 5% of all neonatal admissions). Seventeen percent of hospitals reported 0 outborn admissions. Most infant admissions were of normal BW (median, 59%), followed by low BW (median, 30%) and VLBW (median, 9%). Fifteen hospitals reported data for >1 SAAR-eligible neonatal location. Most neonatal units were in hospitals reporting on just 1 unit or on that unit plus ≥1 level I units (Table 1).

Rates of vancomycin and broad spectrum antibacterial agents predominantly used for hospital-onset infections were associated with the percentage of VLBW admissions (specifically, units in hospitals with a greater proportion of VLBW admissions had greater use, on average) and with level of care (specifically, NICUs in hospitals that accept neonates as transfers for complex procedures had highest use, on average, followed by NICUs in hospitals that do not accept neonates as transfers; level II units had the lowest use) (Table 2, Supplemental Table 5).

Rates of third-generation cephalosporin use differed by facility type, medical school affiliation, and number of hospital beds. Rates of ampicillin and aminoglycoside use, on average, were higher in hospitals reporting no outborn admissions compared with those reporting at least 1 outborn admission and higher in hospitals with a lower proportion of VLBW admissions. On average, rates of fluconazole use were higher in hospitals with a greater proportion of infants weighing <750 g. Usage rates for all antibacterial agents combined were associated with the number of outborn admissions, percentage of annual admissions considered normal BW, total number of admissions, and other levels of care provided (Table 2, Supplemental Table 5).

Compared with NICUs, level II special care nurseries had lower median SAAR values for vancomycin and broad spectrum antibacterial agents predominantly used for hospital-onset infections and higher median SAAR values for ampicillin and aminoglycosides. SAAR distributions for the all antibacterial agents group were similar across location types, although level III (and/or IV) NICUs had a narrower distribution overall (quartile coefficient of dispersion, 0.24 vs 0.36 and 0.37 in level II and II/III units, respectively) (Table 3).

The neonatal SAAR, developed through the collaborative efforts of the CDC and VON, is a new tool for hospital-based neonatal ASPs to use in their efforts to measure and improve AU in a patient population that is frequently exposed to antimicrobial agents. The availability of the neonatal SAAR, coupled with NHSN’s increasing surveillance coverage of AU in neonatal units, enables measurement and improvement activities that extend beyond individual hospitals to a larger community of practice. Our summary of SAAR distributions across neonatal patient care locations in this report is an initial snapshot, with a larger picture on the horizon. As of March 1, 2021, 928 hospitals had voluntarily reported AU data to NHSN from ≥1 neonatal units.

The predictive modeling used to develop neonatal SAARs revealed important differences in AU predictors compared with adult and pediatric populations. Hospital-level variables, such as number of beds, medical school affiliation, and facility type, were less likely to predict AU among neonates. Only the third-generation cephalosporins model included these covariates in the final risk-adjusted SAAR model. Neonatal-specific hospital data were stronger predictors of neonatal AU. Rates of vancomycin and broad spectrum antibacterial agents predominantly used for hospital-onset infections were higher in hospitals caring for neonates with more complex conditions, such as those with a higher proportion of VLBW infants and those accepting neonates as transfers for various complex procedures (eg, omphalocele repair, ventriculoperitoneal shunt, tracheoesophageal fistula/esophageal atresia repair, bowel resection/reanastomosis, meningomyelocele repair, cardiac catheterization). Rates of ampicillin and aminoglycoside use were higher in hospitals caring for neonates with less complex conditions, such as those with only inborn admissions and those with a lower proportion of VLBW infants. Although these high-level indicators of overall neonatal care proved to be strong predictors of neonatal AU, patient-level data would likely improve risk adjustment and the SAARs’ predictive performance.

Recent studies have shown that rates of neonatal AU have decreased in NICUs across the United States.^12,13  However, wide variation persists in neonatal AU across units and hospitals, including variability in the proportion of VLBW infants who receive antibiotics for early-onset sepsis and variability in the duration of antimicrobial treatment.^13–15  Furthermore, correlation between rates of AU and proven infection is often lacking.^13,14  VON partnered with the CDC in 2016 in a multicenter, Internet-based quality improvement (QI) learning collaborative effort, Choosing Antibiotics Wisely, with the goal of reducing overuse and misuse of antibiotics in newborns. The QI program included education and resources to apply QI science to the CDC’s Core Elements of Hospital ASPs.¹⁶  Prospectively planned, cross-sectional, 1-day audits were conducted each year in February and November of 2016 and 2017. The median percentage of infants in the NICU receiving ≥1 antibiotics on the day of the audit decreased from 16.7% to 12.1% (P for trend < .0013), a 34% relative risk reduction.¹⁷  Despite improvement across all participating centers, there was considerable variability with respect to the percentage of NICU infants receiving antibiotics. AU variability is also evident in neonatal SAARs; generally, large quartile coefficients of dispersion are observed with neonatal SAARs, larger than seen with adult and pediatric SAARs.⁹  With such variation in neonatal AU, SAAR distributions help ASPs seeking to understand how AU in their clinical setting compares with use nationally.

Large variation in neonatal AU, reflected here by wide neonatal SAAR distributions and large quartile coefficients of variation, may suggest that some AU is unnecessary.¹⁵  Variation in use is particularly high for broad-spectrum agents predominantly used for hospital-onset infections, third-generation cephalosporins, fluconazole, and vancomycin in level II units (quartile coefficients of dispersion range, 0.70–1.00). Hospitals can use distributions shown here to identify outlying SAAR values, which can prompt them to take a closer look at their prescribing practices and assess whether AU can be improved.

Although in various studies, researchers have observed large variation in neonatal AU that is unrelated to infection rates, we did not assess the relationship between neonatal AU and patient-level risk factors; therefore, we cannot rule out the possibility that some variation in SAARs is due to factors not taken into account in risk models.^13,14  An additional limitation is that because SAAR risk adjustments are limited to location- and facility-level characteristics, we rely on accurate data reporting from hospitals to ensure proper categorization of these characteristics in SAAR models, and we encourage hospitals to validate their location mapping, AU data, and survey data, at least annually.

Antibiotic stewards often use the SAAR to identify patient care locations with AU that deviates from what is predicted, units requiring ASP support, interventions that may be needed, and to track improvements in AU once interventions have been implemented. As ASPs use neonatal SAARs, NHSN anticipates that the ASPs will provide feedback about the benchmark’s strength and limitations. The CDC and VON began hosting quarterly neonatal-specific AU Option user calls in 2020 to encourage sharing of information, to understand how to better support neonatal AU surveillance and ASPs, and to foster a broader understanding of neonatal SAARs and variation in AU. We plan to apply what we learn from users and ASP-SIG members to make improvements in future SAAR models. In 2021, the CDC plans to launch a neonatal component in NHSN where hospitals can report patient-level information on late-onset sepsis and neonatal meningitis. Once implemented, we will assess possible associations between neonatal infection rates and AU at the hospital-level, and we intend to incorporate infection data and other patient-level information into future iterations of neonatal SAARs.

The authors thank all consulted subject matter experts for their assistance with development of SAAR antimicrobial agent categories and their ongoing dedication to antibiotic stewardship.