OBJECTIVES:

We sought to identify whether and how the NICU antibiotic use rate (AUR), clinical correlates, and practice variation changed between 2013 and 2016 and attempted to identify AUR ranges that are consistent with objectively determined bacterial and/or fungal disease burdens.

METHODS:

In a retrospective cohort study of >54 000 neonates annually at >130 California NICUs from 2013 to 2016, we computed nonparametric linear correlation and compared AURs among years using a 2-sample test of proportions. We stratified by level of NICU care and participation in externally organized stewardship efforts.

RESULTS:

By 2016, the overall AUR declined 21.9% (95% confidence interval [CI] 21.9%–22.0%), reflecting 42 960 fewer antibiotic days. Among NICUs in externally organized antibiotic stewardship efforts, the AUR declined 28.7% (95% CI 28.6%–28.8%) compared with 16.2% (95% CI 16.1%–16.2%) among others. The intermediate NICU AUR range narrowed, but the distribution of values did not shift toward lower values as it did for other levels of care. The 2016 AUR correlated neither with proven infection nor necrotizing enterocolitis. The 2016 regional NICU AUR correlated with surgical volume (ρ = 0.53; P = .01), mortality rate (ρ = 0.57; P = .004), and average length of stay (ρ = 0.62; P = .002) and was driven by 3 NICUs with the highest AUR values (30%–57%).

CONCLUSIONS:

Unexplained antibiotic use has declined but continues. Currently measured clinical correlates generally do not help explain AUR values that are above the lowest quartile cutpoint of 14.4%.

What’s Known on This Subject:

Neonatal antibiotic prescribing practice varies widely and is unexplained by the burden of proven infection or other unambiguous clinical indications. Researchers in recent publications and antibiotic stewardship efforts aim to reduce such variation.

What This Study Adds:

Antibiotic use rates are declining; practice variation is narrowing yet still largely unexplained. Our findings help inform a framework for estimating antibiotic use rate ranges that are consistent with objectively determined bacterial and/or fungal disease burdens.

Recent work reveals that a considerable portion of NICU antibiotic use is unwarranted.1,2 In a 2013 sample of 52 061 infants receiving 746 051 patient-days of NICU care, antibiotic use rates (AURs) varied widely across NICUs and were unexplained by the corresponding burden of proven infection or other clearly warranted indications for treatment.1 Antibiotic prescribing practice variation apparently hinges on variation in how practitioners frame, interpret, and respond to clinical situations that are ultimately considered unproven infection. NICU staff differ in their indications for treatment3,4 and conceptualizations of “culture-negative sepsis.”2 

Such practice variation plausibly may be mitigated with the aid of carefully conceived antibiotic stewardship initiatives used to promote current evidence-based practice.5 During the past 2 years, the Vermont Oxford Network partnered with the Centers for Disease Control and Prevention in a collaborative involving 167 NICUs in 38 states, Puerto Rico, and 6 countries.6 The California Perinatal Quality Care Collaborative (CPQCC)7,8 began a collaborative involving 28 California NICUs along with an alternative entailing similar interventions minus facilitated access to the other centers participating in the collaborative or its expert panel.

Accurately estimating the impact of such work requires understanding concurrent trends: are changes in measured outcomes restricted to NICUs engaged with specified interventions, or are outcomes changing more broadly? For example, 1 large NICU recently reported a 27% decrease in days of therapy per 1000 patient-days after a local antibiotic stewardship intervention.9 Recently published recommendations may also be driving practice changes.3,4,10,12 Individual NICU improvement efforts may be informal and may not be reported in peer-reviewed literature. Informed clinical, payer, and regulatory evaluations of resource use and an unbiased interpretation of formal improvement efforts require measuring and understanding such concurrent trends.

Since 2013, the California Department of Health Care Services has tracked NICU antibiotic use at all NICUs approved by California Children’s Services (CCS); additional NICUs have also reported these data to the CPQCC. In the current study, we seek to identify (1) changes over time in NICU antibiotic use and the range of practice variation, (2) whether concurrent trends differed from performance at NICUs participating in known externally organized antibiotic stewardship efforts (described above), and (3) whether clinical and resource use correlations with AUR changed since 2013.1 

These insights are essential for estimating AUR ranges that are consistent with objectively determined bacterial and/or fungal disease burdens, a topic that has not yet been addressed in the literature. The stewardship initiatives and publications cited above reflect a broad-based view that NICU AURs often are unnecessarily high, begging the question, “Which rate is right?”

CCS confers state approval for 3 levels of NICU care: regional, community, and intermediate.13 These generally correspond to American Academy of Pediatrics levels IV, III, and II, respectively.14 CCS standards include required annual data reporting of specific variables. All CCS-approved NICUs submit their data to the CPQCC,15 which prepares an annual report for each NICU and submits an aggregate data set to CCS. The CPQCC collects clinical data prospectively using an expanded version of the Vermont Oxford Dataset16,17 and a supplemental CCS form. All CCS-approved NICUs must complete the supplemental CCS form; the CPQCC requires that other NICUs also do so. Non–CCS-approved NICUs may be level II or III and typically choose not to apply for approval. The collected data constitute a single overarching database.

Of the 148 NICUs in California,18 137 currently belong to the CPQCC, and 122 are CCS approved. Thus, the combined data set is used to describe NICU care and outcomes for most neonates receiving NICU care in California. For the study analysis, we used CPQCC–CCS data sets for calendar years 2013 to 2016. This study was approved by the Stanford University Institutional Review Board.

All neonates receiving care at CPQCC–CCS NICUs in California reporting on data variables used for this study were included. The number of NICUs varied from 131 to 137 by year (Table 1).

TABLE 1

NICU Patient Volume and Antibiotic Use Stratified by Participation in Externally Organized Antibiotic Stewardship Efforts

No. NICUsNo. AdmissionsNo. Patient dNo. Antibiotic dOverall AUR, %AUR Median/Mean (SD)Range of AUR, Multiples of Lowest to Highest ValueLowest/Highest AUR Value, %
2013         
 All NICUs 131 54 181 777 194 214 370 27.6 24.2/28 (17.8) >100 0.01/97.1 
 Regional 22 
 Community 82 
 Intermediate 14 
 Non-CCS 13 
 Collaboratives 43 22 936 355 839 99 613 28.0 26.4/29.4 (17.9) >100 0.01/89.8 
 Nonparticipants 88 31 245 421 355 114 757 27.2 23.8/27.3 (17.8) >100 0.2/97.1 
2014         
 All NICUs 131 56 146 796 258 215 355 27.0 23.8/27.2 (15.4) 78 1.1/84.9 
 Regional 23 
 Community 81 
 Intermediate 13 
 Non-CCS 14 
 Collaboratives 44 23 386 361 102 98 782 27.4 26.2/28.6 (14.1) 13 6.3/80.4 
 Nonparticipants 87 32 760 435 156 116 573 26.8 22.3/26.5 (16.1) 77 1.1/85 
2015         
 All NICUs 134 57 332 791 220 195 679 24.7 21.4/24.4 (12.2) 7.3/66.7 
 Regional 23 
 Community 81 
 Intermediate 14 
 Non-CCS 16 
 Collaboratives 42 23 294 341 260 80 389 23.6 21.9/24 (10.4) 7.3/52 
 Nonparticipants 92 34 038 449 960 115 290 25.6 21/24.6 (13) 8.2/66.7 
2016         
 All NICUs 137 56 257 795 663 171 410 21.6 19.7/21.8 (10) 10 5.6/57 
 Regional 23 
 Community 83 
 Intermediate 17 
 Non-CCS 14 
 Collaboratives 44 23 127 356 822 71 200 19.9 19/21.1 (8.2) 9.4/46.8 
 Nonparticipants 93 33 130 438 841 100 210 22.8 20.1/22.2 (10.7) 10 5.6/57 
No. NICUsNo. AdmissionsNo. Patient dNo. Antibiotic dOverall AUR, %AUR Median/Mean (SD)Range of AUR, Multiples of Lowest to Highest ValueLowest/Highest AUR Value, %
2013         
 All NICUs 131 54 181 777 194 214 370 27.6 24.2/28 (17.8) >100 0.01/97.1 
 Regional 22 
 Community 82 
 Intermediate 14 
 Non-CCS 13 
 Collaboratives 43 22 936 355 839 99 613 28.0 26.4/29.4 (17.9) >100 0.01/89.8 
 Nonparticipants 88 31 245 421 355 114 757 27.2 23.8/27.3 (17.8) >100 0.2/97.1 
2014         
 All NICUs 131 56 146 796 258 215 355 27.0 23.8/27.2 (15.4) 78 1.1/84.9 
 Regional 23 
 Community 81 
 Intermediate 13 
 Non-CCS 14 
 Collaboratives 44 23 386 361 102 98 782 27.4 26.2/28.6 (14.1) 13 6.3/80.4 
 Nonparticipants 87 32 760 435 156 116 573 26.8 22.3/26.5 (16.1) 77 1.1/85 
2015         
 All NICUs 134 57 332 791 220 195 679 24.7 21.4/24.4 (12.2) 7.3/66.7 
 Regional 23 
 Community 81 
 Intermediate 14 
 Non-CCS 16 
 Collaboratives 42 23 294 341 260 80 389 23.6 21.9/24 (10.4) 7.3/52 
 Nonparticipants 92 34 038 449 960 115 290 25.6 21/24.6 (13) 8.2/66.7 
2016         
 All NICUs 137 56 257 795 663 171 410 21.6 19.7/21.8 (10) 10 5.6/57 
 Regional 23 
 Community 83 
 Intermediate 17 
 Non-CCS 14 
 Collaboratives 44 23 127 356 822 71 200 19.9 19/21.1 (8.2) 9.4/46.8 
 Nonparticipants 93 33 130 438 841 100 210 22.8 20.1/22.2 (10.7) 10 5.6/57 

The AUR is the number of patient-days that infants were exposed to ≥1 antibacterial or antifungal agents administered intravenously or intramuscularly per 100 patient-days in the reporting NICU expressed as a percentage. The early-onset sepsis (EOS) rate is the percentage of infants with a bacterial or fungal infection diagnosed by blood culture within 2 days of birth. The central line–associated bloodstream infection rate is the number of laboratory-confirmed bloodstream infections for which a central line was in place for 2 days on the date of the event per 1000 central line–days. The inborn admission rate is the proportion of all live births at a hospital who were admitted to the NICU. The number of surgical cases includes patients undergoing a surgical procedure, excluding circumcision, extracorporeal membrane oxygenation cannulation and/or decannulation, peritoneal dialysis catheter placement and/or removal, chest tube placement, and central line placement. The NICU mortality rate is the ratio of NICU deaths to the number of NICU admissions. The inborn admission rate is the proportion of live births at a hospital who were admitted to the NICU expressed as a percent value. Twelve reporting hospitals were self-designated free-standing NICUs that are organizationally independent of a maternal delivery service. Data used to quantify the service birth population (located in the same building as the NICU) were available for 8 of these and were used to compute inborn admission rates.

Certain CPQCC–CCS variables are restricted to infants who were 401 g to 1500 g or 22 to 29 weeks’ gestation at birth. The rate of late-onset sepsis (LOS) is the percentage of infants with a bacterial or fungal infection diagnosed by blood culture at ≥3 days after birth. The fungal infection rate is the percentage of infants with a fungal infection diagnosed by blood culture at ≥3 days after birth. The necrotizing enterocolitis (NEC) rate is the percentage of infants diagnosed with NEC either at surgery, at postmortem examination, or by radiographic pneumatosis intestinalis, hepatobiliary gas, and/or pneumoperitoneum accompanied by bilious gastric aspirate or emesis, abdominal distention, and/or gross or occult blood in stool with no apparent rectal fissure. The average length of stay is the average NICU length of stay in days for patients discharged from the hospital.

Study variable values were computed by calendar year.

The unit of observation and analysis was the individual NICU. The primary outcome was the AUR. Before the study began, we formulated hypotheses that were consistent with goals for NICU antibiotic stewardship efforts6: the AUR would correlate positively with the burden of proven infection, with NEC, with the number of surgical cases, and with highest NICU level of care. We examined post hoc (1) the correlation between the number of surgical cases and NICU mortality rate; (2) 4-year differences in intermediate NICU admissions, patient-days, and antibiotic use–days; (3) the mean clinical and resource correlate values for NICUs in AUR quartile 1 versus quartiles 2 to 4.

We estimated linear correlation by Spearman rank correlation to avoid reliance on specific distributional assumptions and mitigate effects of extreme outlier values. We compared the AUR for different time periods by 2-sample tests of proportions. We compared AUR correlate sample means by analysis of variance; when variances were unequal by Bartlett’s test, we used 2-sample t tests with unequal variances. We compared differences in intermediate NICU admissions, patient-days, and antibiotic-days over 2, 3, and 4 years by paired t tests. Hypothesis tests were 2-sided with a significance level at P ≤ .05. We used stratified analyses to examine the NICU level of care or participation in known externally organized antibiotic stewardship efforts. We used Stata 15 (Stata Corp, College Station, TX)19 for analyses and graphical displays.

As detailed in Table 1 and Fig 1, by 2016, the overall AUR declined by 21.9% (95% confidence interval [CI] 21.9%–22.0%), and AUR variation narrowed among NICUs by fivefold to 10-fold. Antibiotic-days declined by 42 960 (from 214 370 to 171 410) despite 18 469 additional patient-days (from 777 194 to 795 663). The 44 NICUs participating in known externally organized antibiotic stewardship efforts accounted for 41% of NICU admissions, 41% of antibiotic-days, and 45% of patient-days. The 2013 AUR among NICUs later participating in known externally organized antibiotic stewardship efforts was 28.0% and 27.2% among those that did not (difference of 0.8%; 95% CI 0.6%–1.0%). The 2016 AUR among NICUs in known externally organized antibiotic stewardship efforts declined to 19.9% (28.7% decrease; 95% CI 28.6%–28.8%) compared with 22.8% (16.2% decrease; 95% CI 16.1%–16.2%) among NICUs that did not participate (19.9% vs 22.8%; 12.7% relative difference; 95% CI 12.5%–12.7%).

FIGURE 1

AUR ranges for all NICUs, those not participating in externally organized antibiotic stewardship efforts, and those participating in externally organized antibiotic stewardship efforts. Shaded rectangles indicate the interquartile range and median. Lines above or below the box extend farther by 1.5 times the interquartile range. Dots indicate extreme outliers. A, 2013. B, 2014. C, 2015. D, 2016. AS, externally organized antibiotic stewardship efforts.

FIGURE 1

AUR ranges for all NICUs, those not participating in externally organized antibiotic stewardship efforts, and those participating in externally organized antibiotic stewardship efforts. Shaded rectangles indicate the interquartile range and median. Lines above or below the box extend farther by 1.5 times the interquartile range. Dots indicate extreme outliers. A, 2013. B, 2014. C, 2015. D, 2016. AS, externally organized antibiotic stewardship efforts.

Close modal

Figure 2 illustrates the range of AUR values and relative frequency of occurrence stratified by NICU level of care. For regional, community, and non-CCS NICUs, the AUR range narrowed and the bulk of the distributions shifted toward lower values. The AUR distribution pattern for intermediate NICUs noticeably differs: although the range narrowed, the bulk of the distribution of values did not shift toward lower values, as it did at other levels of care. Intermediate NICU admissions, patient-days, and antibiotic-days did not significantly differ across years.

FIGURE 2

Distribution of AUR values by level of care for all NICUs from 2013 to 2016. The y-axis displays kernel density, which is essentially a smoothed frequency distribution histogram used to estimate the density of the distribution of values (the relative percent of NICUs at each admission rate value). A, 2013. B, 2014. C, 2015. D, 2016.

FIGURE 2

Distribution of AUR values by level of care for all NICUs from 2013 to 2016. The y-axis displays kernel density, which is essentially a smoothed frequency distribution histogram used to estimate the density of the distribution of values (the relative percent of NICUs at each admission rate value). A, 2013. B, 2014. C, 2015. D, 2016.

Close modal

The 2016 AUR correlated neither with proven infection nor NEC, echoing our 2013 findings1 (Table 2). Only among regional NICUs did the 2016 AUR correlate with surgical case volume (ρ = 0.53; P = .01), NICU mortality rate (ρ = 0.57; P = .004), and average length of stay (ρ = 0.62; P = .002). However, among regional NICUs in quartile 1 (≤14.4%), the AUR did not correlate so. The AUR correlated with inborn admission rate across all NICUs (ρ = 0.23; P = .009) and community NICUs (ρ = 0.3; P = .006). The latter correlation did not persist when examined for community NICUs in AUR quartile 1 (≤14.3%) or quartiles 2 to 4. Proven infection rates at NICUs in AUR quartiles 2 to 4 were statistically similar at NICUs in quartile 1. The NEC rate was significantly higher among community NICUs in AUR quartiles 2 to 4 compared with quartile 1 (absolute rate difference 1.11%; P = .03). Surgical volume was higher among regional NICUs in AUR quartiles 2 to 4 (mean 243.94; SD 159.43) compared with quartile 1 (mean 76.80; SD 69.92; P = .03).

TABLE 2

Estimated Correlations With AURs and Correlate Variable Sample Means for Quartiles 1 Versus Quartiles 2–4

Correlation Coefficient, ρPAUR Quartile 1, Mean (SD)AUR Quartiles 2–4, Mean (SD)P
EOS      
 All NICUs 0.10 .29 0.59 (0.74) 0.61 (0.64) .86 
 Regional 0.02 .93 0.47 (0.46) 0.73 (0.81) .51 
 Community 0.11 .33 0.66 (0.83) 0.61 (0.57) .78 
 Intermediate 0.20 .47 0.25 (0.43) 0.45 (0.69) .64 
 Non-CCS a N/A b b N/A 
CLABSI      
 All NICUs −0.02 .81 0.81 (2.19) 0.98 (1.59) .69 
 Regional 0.27 .21 0.91 (0.96) 1.25 (0.68) .78 
 Community −0.08 .45 1.07 (2.71) 1.07 (1.83) .99 
 Intermediate a N/A a a N/A 
 Non-CCS 0.40 .20 0 (0) 1.38 (1.94) .19 
LOS      
 All 0.03 .77 2.76 (2.94) 3.18 (2.98) .48 
 Regional 0.20 .37 3.14 (1.35) 3.98 (2.17) .43 
 Community −0.07 .50 3.40 (3.29) 3.19 (2.79) .79 
 Intermediate 0.33 .19 0 (0) 1.10 (2.49) .46 
 Non-CCS 0.36 .20 1.38 (2.19) 4.67 (4.68) .17 
Fungal infection      
 All 0.05 .60 0.08 (0.28) 0.19 (0.57) .14 
 Regional −0.18 .41 0.57 (0.56) 0.31 (0.46) .30 
 Community 0.15 .17 0 (0) 0.18 (0.57) .15 
 Intermediate a N/A a a N/A 
 Non-CCS 0.31 .28 0 (0) 0.35 (1.04) .48 
NEC      
 All NICUs 0.01 .88 1.47 (2.27) 2.23 (2.89) .16 
 Regional 0.04 .87 4.59 (1.77) 5.03 (3.02) .76 
 Community 0.11 .34 0.83 (1.71) 1.94 (2.64) .03 
 Intermediate a N/A a a N/A 
 Non-CCS −0.003 .99 1.90 (2.93) 1.56 (2.33) .82 
No. surgical cases      
 All NICUs −0.04 .67 14.97 (37.28) 50.07 (112.49) .007 
 Regional 0.53 .01 76.80 (69.92) 243.94 (159.43) .03 
 Community 0.07 .52 5.48 (13.75) 11.97 (24.44) .14 
 Intermediate −0.17 .51 1.33 (2.31) 0.14 (0.53) .47 
 Non-CCS −0.37 .19 1.20 (1.09) 2.44 (5.50) .53 
NICU mortality rate      
 All NICUs −0.08 .34 0.01 (0.01) 0.01 (0.01) .69 
 Regional 0.57 .004 0.02 (0.01) 0.03 (0.02) .13 
 Community −0.13 .23 0.01 (0.01) 0.01 (0.01) .39 
 Intermediate 0.10 .70 0 (0) 0.001 (0.003) .66 
 Non-CCS 0.08 .78 0.002 (0.003) 0.005 (0.005) .28 
Inborn admission rate      
 All NICUs 0.23 .009 11.20 (7.12) 13.03 (5.35) .18 
 Regional 0.26 .30 16.90 (15.32) 18.73 (7.30) .81 
 Community 0.30 .006 10.92 (4.92) 12.66 (4.72) .17 
 Intermediate 0.07 .79 9.54 (1.80) 11.45 (3.60) .23 
 Non-CCS 0.39 .16 7.68 (1.44) 9.85 (2.41) .06 
Average length of stay      
 All NICUs 0.05 .56 55.06 (12.87) 58.69 (21.69) .25 
 Regional 0.62 .002 66.84 (8.47) 82.87 (19.02) .08 
 Community 0.06 .56 56.81 (10.36) 59.89 (11.76) .30 
 Intermediate −0.29 .27 36.44 (11.74) 21.91 (13.95) .12 
 Non-CCS 0.53 .06 47.47 (12.42) 54.70 (16.03) .41 
Correlation Coefficient, ρPAUR Quartile 1, Mean (SD)AUR Quartiles 2–4, Mean (SD)P
EOS      
 All NICUs 0.10 .29 0.59 (0.74) 0.61 (0.64) .86 
 Regional 0.02 .93 0.47 (0.46) 0.73 (0.81) .51 
 Community 0.11 .33 0.66 (0.83) 0.61 (0.57) .78 
 Intermediate 0.20 .47 0.25 (0.43) 0.45 (0.69) .64 
 Non-CCS a N/A b b N/A 
CLABSI      
 All NICUs −0.02 .81 0.81 (2.19) 0.98 (1.59) .69 
 Regional 0.27 .21 0.91 (0.96) 1.25 (0.68) .78 
 Community −0.08 .45 1.07 (2.71) 1.07 (1.83) .99 
 Intermediate a N/A a a N/A 
 Non-CCS 0.40 .20 0 (0) 1.38 (1.94) .19 
LOS      
 All 0.03 .77 2.76 (2.94) 3.18 (2.98) .48 
 Regional 0.20 .37 3.14 (1.35) 3.98 (2.17) .43 
 Community −0.07 .50 3.40 (3.29) 3.19 (2.79) .79 
 Intermediate 0.33 .19 0 (0) 1.10 (2.49) .46 
 Non-CCS 0.36 .20 1.38 (2.19) 4.67 (4.68) .17 
Fungal infection      
 All 0.05 .60 0.08 (0.28) 0.19 (0.57) .14 
 Regional −0.18 .41 0.57 (0.56) 0.31 (0.46) .30 
 Community 0.15 .17 0 (0) 0.18 (0.57) .15 
 Intermediate a N/A a a N/A 
 Non-CCS 0.31 .28 0 (0) 0.35 (1.04) .48 
NEC      
 All NICUs 0.01 .88 1.47 (2.27) 2.23 (2.89) .16 
 Regional 0.04 .87 4.59 (1.77) 5.03 (3.02) .76 
 Community 0.11 .34 0.83 (1.71) 1.94 (2.64) .03 
 Intermediate a N/A a a N/A 
 Non-CCS −0.003 .99 1.90 (2.93) 1.56 (2.33) .82 
No. surgical cases      
 All NICUs −0.04 .67 14.97 (37.28) 50.07 (112.49) .007 
 Regional 0.53 .01 76.80 (69.92) 243.94 (159.43) .03 
 Community 0.07 .52 5.48 (13.75) 11.97 (24.44) .14 
 Intermediate −0.17 .51 1.33 (2.31) 0.14 (0.53) .47 
 Non-CCS −0.37 .19 1.20 (1.09) 2.44 (5.50) .53 
NICU mortality rate      
 All NICUs −0.08 .34 0.01 (0.01) 0.01 (0.01) .69 
 Regional 0.57 .004 0.02 (0.01) 0.03 (0.02) .13 
 Community −0.13 .23 0.01 (0.01) 0.01 (0.01) .39 
 Intermediate 0.10 .70 0 (0) 0.001 (0.003) .66 
 Non-CCS 0.08 .78 0.002 (0.003) 0.005 (0.005) .28 
Inborn admission rate      
 All NICUs 0.23 .009 11.20 (7.12) 13.03 (5.35) .18 
 Regional 0.26 .30 16.90 (15.32) 18.73 (7.30) .81 
 Community 0.30 .006 10.92 (4.92) 12.66 (4.72) .17 
 Intermediate 0.07 .79 9.54 (1.80) 11.45 (3.60) .23 
 Non-CCS 0.39 .16 7.68 (1.44) 9.85 (2.41) .06 
Average length of stay      
 All NICUs 0.05 .56 55.06 (12.87) 58.69 (21.69) .25 
 Regional 0.62 .002 66.84 (8.47) 82.87 (19.02) .08 
 Community 0.06 .56 56.81 (10.36) 59.89 (11.76) .30 
 Intermediate −0.29 .27 36.44 (11.74) 21.91 (13.95) .12 
 Non-CCS 0.53 .06 47.47 (12.42) 54.70 (16.03) .41 

From the 2016 data set. CLABSI, central line–associated bloodstream infection; N/A, not available; —, not applicable.

a

NICUs reported a rate of 0.

b

Not reported.

From 2013 to 2016, the overall AUR declined by 21.9%; antibiotic exposure–days declined by ∼43 000. The continuing lack of correlation between AURs and proven infection or NEC reveals that at many NICUs, the AUR can decrease further. Among NICUs participating in known externally organized antibiotic stewardship efforts, the AUR declined by 28.7%; among the other NICUs, the AUR declined by 16.2%. These differences are statistically and clinically highly significant (tens of thousands of antibiotic exposure–days avoided). Although the 2013 baseline of 0.8% absolute AUR difference between these groups (Table 1) was statistically significant (reflecting large numbers of patient- and antibiotic-days), clinical significance may be debatable. By 2015, the overall AUR had already declined by 9.7% (95% CI 9.3%–10.1%). California NICUs began to receive AUR feedback in mid-2014; 1 pertinent guideline reappraisal was published in 2015.3 We speculate that 2015 AUR values began to reflect new publications1,3,5 and the initiation of the antibiotic stewardship efforts described above.

Our correlation analyses rest on 2 key assumptions: (1) antibiotic exposures should correlate with objectively measured bacterial and/or fungal disease burden, and (2) the selected correlates represent most of the objectively measured NICU bacterial disease burden. Therefore, 2 possible explanations exist for the absence of AUR correlations: (1) ≥1 objectively determined bacterial diseases that are quantitatively important antibiotic use drivers were omitted from the analysis, or (2) AUR variation across California NICUs is unrelated to the commonly occurring indications for treatment that are consistent with antibiotic stewardship principles. Our data do not enable an estimate of a NICU’s burden of blood culture–negative bacterial disease justifying antibiotic treatment (eg, meningitis, omphalitis, osteomyelitis, tracheitis, and pneumonia). Although in our experience such blood culture–negative conditions are not likely to be important AUR drivers at most NICUs, we encourage NICUs to examine in detail this aspect of the AUR.

Considered in isolation, the positive correlations at regional NICUs between the AUR and number of surgical cases (ρ = 0.53; P = .01), NICU mortality rate (ρ = 0.57; P = .004), and average length of stay (ρ = 0.62; P = .002) might reveal that at regional NICUs, the AUR represents a proxy measure of surgical case volume and illness severity. However, Fig 3A illustrates a strong positive correlation between surgical volume and the NICU mortality rate (ρ = 0.74; P < .001). Figures 3C and 3D show that the correlation between the AUR and each of these 2 variables does not scale linearly. The correlations are largely driven by only 3 of 23 NICUs, those with the highest AUR values (30%–57%). The data points indicating the other 20 regional NICUs are contained within the dashed rectangles and show no significant correlation. Interestingly, Fig 3B illustrates that the surgical case volume and mortality relationship largely persists among the 20 regional NICUs with an AUR <30% (ρ = 0.68; P = .001) as well as when the AUR and average length of stay are similarly evaluated (ρ = 0.47; P = .04). Overall, the evidence reveals that although surgical case volume and illness severity might sometimes drive the AUR, a relatively high AUR also may be used to independently predict mortality and average length of stay. In a recent report, the Canadian Neonatal Network found that higher AURs were associated with adverse neonatal outcomes among infants without culture-proven sepsis or NEC, including a twofold increase in mortality odds associated with a 10% AUR increase.20 A possible contributing factor is dysbiosis associated with neonatal antibiotic exposure.21,22 It is thus particularly noteworthy that regional NICU AURs did not correlate with culture-proven sepsis or NEC. Figure 3 thus illustrates opportunities for antibiotic stewardship efforts used to target NICU surgical practice.

FIGURE 3

A, Regional NICUs, all AUR values (ρ = 0.75; P < .001). B, Regional NICUs with AUR <30% (ρ = 0.68; P = .001). C, Regional NICUs (ρ = 0.53; P = .01). D, Regional NICUs (ρ = 0.57; P = .004). Centers within dashed rectangles have an AUR <30%. Straight lines indicate fitted values.

FIGURE 3

A, Regional NICUs, all AUR values (ρ = 0.75; P < .001). B, Regional NICUs with AUR <30% (ρ = 0.68; P = .001). C, Regional NICUs (ρ = 0.53; P = .01). D, Regional NICUs (ρ = 0.57; P = .004). Centers within dashed rectangles have an AUR <30%. Straight lines indicate fitted values.

Close modal

When the 2013 data were initially reported, several NICUs reporting an AUR value <1% were dropped from the analysis because the values were considered clinically implausible.1 To explore the trajectory of these NICU values over time, no censoring was applied in the current study. NICU data accuracy may have improved with increasing AUR collection experience and use of database technology. Since 2015, the lowest reported AUR was 5.6%. In 2013, the cutpoint value for the lowest AUR quartile was 17.2%; in 2016, it was 14.4%.

Our findings can inform a framework for estimating AUR ranges that are consistent with objectively determined bacterial and/or fungal disease burdens. The 2016 cutpoint value for the lowest regional NICU AUR quartile was the same as for all NICUs combined: 14.4%. Among these low-quartile regional NICUs, the AUR did not correlate with surgical volume, NICU mortality rate, or any other variable, including proven infection or NEC. However, these same low-quartile NICUs have an average lower surgical volume than all regional NICUs. Thus, complex relationships remain to be disentangled, and not only for higher-volume surgical centers. A correlation between the AUR and inborn admission rate for community-level NICUs may similarly reflect unaccounted bias and/or confounding; it vanished when examined solely for community NICUs in AUR quartile 1 or for community NICUs in quartiles 2 to 4. Interpretation of higher NEC rates at community NICUs (and not at regional NICUs) in AUR quartiles 2 to 4 (Table 2) is constrained by a lack of patient-level data, including exposure and/or outcome timing and transfer-in after NEC onset. The 1.11% absolute difference in the NEC rate negligibly accounts for the higher AURs.

Interestingly, some NICUs already in the lowest 2013 AUR quartile nonetheless believed there was further opportunity to improve, joining stewardship collaboratives in subsequent years; their belief is confirmed by the even lower 2016 quartile 1 value and narrowed variation (Fig 1). The distribution of AUR values among intermediate-level NICUs thus raises particular concern. The 2016 cutpoint value for the lowest AUR quartile among intermediate NICUs was 17.5%, and the overall distribution shifts toward relatively higher values. Why might intermediate NICUs operate with higher AURs than other NICUs (Fig 2)? All else remaining the same, if decision rules at intermediate NICUs changed over time so that fewer neonates were admitted to rule out sepsis, then patient-days could decrease disproportionately to antibiotic-days. Ironically, the AUR could rise. However, the numbers of intermediate NICU admissions, patient-days, and antibiotic-days did not significantly differ across years, nor were EOS rates higher than at other levels of NICUs (Table 2). Therefore, antibiotic overuse is clearest at intermediate NICUs with both a relatively high AUR and inborn admission rate, and this may support other observations23,24 that reveal a component of supply-sensitive care.25,26 Supply-sensitive care has little evidence base, reflecting instead available service capacity and payment systems that are used to incentivize resource use.26 In other words, the availability of NICU beds becomes a determinant of NICU care.27 

The AUR, as a single measure of NICU-level antibiotic use, necessarily entails imprecisely answering, “Which rate is right?” A NICU’s AUR is used to summarize aggregated results of several different care processes and subpopulations with varying risk profiles (eg, asymptomatic term and near-term neonates with suspected EOS, extremely preterm neonates with suspected EOS or LOS,28 neonates undergoing surgical procedures [again, a heterogeneous subpopulation], and [as already mentioned] neonates requiring antibiotic treatment without a positive blood culture result, such as for meningitis or pneumonia). Accounting for such complexity requires further analytical stratification and individual patient-level units of observation and analysis. Calibrating patient antibiotic exposures with treatment indication could provide important complementary insight and requires collecting additional data elements (eg, estimated number needed to treat for each case of EOS or blood culture–negative meningitis).

The persistent lack of association between AURs and currently collected clinical correlates signals a likely opportunity to continue decreasing antibiotic use and the need for a more complete case-mix characterization. With the possible exception of high–surgical volume NICUs, NICUs in AUR quartiles 2 to 4 generally serve populations with similar measured clinical correlates as NICUs in quartile 1 at the same level of care. Thus, for most or all NICUs, currently available data elements reveal no explanation for AURs above the lowest quartile cutpoint of 14.4%. We suggest that all NICUs, and especially those with an AUR beyond quartile 1, interpret their AURs in light of answers to the following systematic critical review: Are our numerator and denominator values accurate? What fraction of our numerator reflects patients with objectively diagnosed bacterial disease not measured by the currently collected data elements? Does our denominator reflect a relatively low inborn admission rate; has it changed over time? And might there be genuine opportunities to improve antibiotic use (eg, a large portion of the numerator represents patients in whom objective bacterial disease cannot be established by current evidence-based methods)?

     
  • AUR

    antibiotic use rate

  •  
  • CCS

    California Children’s Services

  •  
  • CI

    confidence interval

  •  
  • CPQCC

    California Perinatal Quality Care Collaborative

  •  
  • EOS

    early-onset sepsis

  •  
  • LOS

    late-onset sepsis

  •  
  • NEC

    necrotizing enterocolitis

Dr Schulman conceptualized and designed the study, performed the analyses, drafted the initial manuscript, and reviewed and revised the manuscript; Drs Profit, Lee, and Gould supervised data collection and critically reviewed the manuscript; Ms Dueñas, Dr Bennett, and Ms Parucha coordinated and supervised data collection, aggregated data, provided the aggregated data sets to California Children’s Services, and critically reviewed the manuscript; Dr Jocson participated in the study design and interpretation of analyses and critically reviewed the manuscript; and all authors approved the final manuscript as submitted and agree to be accountable for all aspects of the work.

FUNDING: No specific support was provided. Dr Profit was supported in part by grants from the Eunice Kennedy Shriver National Institute of Child Health and Human Development (R01 HD083368-01 and R01 HD08467-01; principal investigator: Dr Profit). The content is solely the responsibility of the authors and does not necessarily represent the official views of the Eunice Kennedy Shriver National Institute of Child Health and Human Development or the National Institutes of Health. This funding source had no role in any of the following aspects of this study: design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the article; and decision to submit the article for publication.

1
Schulman
J
,
Dimand
RJ
,
Lee
HC
,
Duenas
GV
,
Bennett
MV
,
Gould
JB
.
Neonatal intensive care unit antibiotic use.
Pediatrics
.
2015
;
135
(
5
):
826
833
[PubMed]
2
Cantey
JB
,
Baird
SD
.
Ending the culture of culture-negative sepsis in the neonatal ICU.
Pediatrics
.
2017
;
140
(
4
):
e20170044
3
Benitz
WE
,
Wynn
JL
,
Polin
RA
.
Reappraisal of guidelines for management of neonates with suspected early-onset sepsis.
J Pediatr
.
2015
;
166
(
4
):
1070
1074
[PubMed]
4
Kuzniewicz
MW
,
Puopolo
KM
,
Fischer
A
, et al
.
A quantitative, risk-based approach to the management of neonatal early-onset sepsis.
JAMA Pediatr
.
2017
;
171
(
4
):
365
371
[PubMed]
5
Soll
RF
,
Edwards
WH
.
Antibiotic use in neonatal intensive care.
Pediatrics
.
2015
;
135
(
5
):
928
929
[PubMed]
6
Vermont Oxford Network
. iNICQ 2018: choosing antibiotics wisely. 2018. Available at: https://public.vtoxford.org/quality-education/inicq-2018/. Accessed September 9, 2017
7
California Perinatal Quality Care Collaborative
. Antibiotic stewardship collaborative. 2017. Available at: https://www.cpqcc.org/qi-projects/cpqccccs-collaboratives/antiobiotic-stewardship-collaborative-5. Accessed September 9, 2017
8
California Perinatal Quality Care Collaborative
. NICU QI. 2017. Available at: https://www.cpqcc.org/qi-projects/nicu-qi. Accessed September 9, 2017
9
Cantey
JB
,
Wozniak
PS
,
Pruszynski
JE
,
Sánchez
PJ
.
Reducing unnecessary antibiotic use in the neonatal intensive care unit (SCOUT): a prospective interrupted time-series study.
Lancet Infect Dis
.
2016
;
16
(
10
):
1178
1184
[PubMed]
10
Committee on Obstetric Practice
.
Committee opinion no. 712: intrapartum management of intraamniotic infection.
Obstet Gynecol
.
2017
;
130
(
2
):
e95
e101
[PubMed]
11
Hooven
TA
,
Polin
RA
.
Time to overhaul the “rule out sepsis” workup.
Pediatrics
.
2017
;
140
(
1
):
e20171155
[PubMed]
12
Higgins
RD
,
Saade
G
,
Polin
RA
, et al;
Chorioamnionitis Workshop Participants
.
Evaluation and management of women and newborns with a maternal diagnosis of chorioamnionitis: summary of a workshop.
Obstet Gynecol
.
2016
;
127
(
3
):
426
436
[PubMed]
13
California Department of Health Care Services
. Provider standards. 2014. Available at: www.dhcs.ca.gov/services/ccs/Pages/ProviderStandards.aspx#nicu. Accessed July 17, 2014
14
American Academy of Pediatrics Committee on Fetus and Newborn
.
Levels of neonatal care.
Pediatrics
.
2012
;
130
(
3
):
587
597
[PubMed]
15
California Perinatal Quality Care Collaborative
. California Perinatal Quality Care Collaborative (CPQCC). Available at: https://www.cpqcc.org/. Accessed April 25, 2018
16
California Perinatal Quality Care Collaborative
. CPQCC data center. 2018. Available at: https://www.cpqcc.org/perinatal-programs/cpqcc-data-center. Accessed April 25, 2018
17
Vermont Oxford Network
. Vermont Oxford Network database: manual of operations for infants born in 2015. 2015. Available at: https://public.vtoxford.org/wp-content/uploads/2014/11/Manual_of_Operations_Part2_v19.pdf. Accessed October 5, 2017
18
Bhatt
DR
,
Adams
M
,
Clearly
J
, et al
.
NICUs & Neonatologists in California
.
California Association of Neonatologists, District IX Section on Neonatal-Perinatal Medicine, California Perinatal Quality Care Collaborative, California Children’s Services
;
2017
19
Stata Statistical Software
[computer program]. Version 15.
College Station, TX
:
Stata Press
;
2017
20
Ting
JY
,
Synnes
A
,
Roberts
A
, et al;
Canadian Neonatal Network Investigators
.
Association between antibiotic use and neonatal mortality and morbidities in very low-birth-weight infants without culture-proven sepsis or necrotizing enterocolitis.
JAMA Pediatr
.
2016
;
170
(
12
):
1181
1187
[PubMed]
21
Mukhopadhyay
S
,
Puopolo
KM
.
Antibiotic use and mortality among premature infants without confirmed infection-perpetrator or innocent bystander?
JAMA Pediatr
.
2016
;
170
(
12
):
1144
1146
[PubMed]
22
Vangay
P
,
Ward
T
,
Gerber
JS
,
Knights
D
.
Antibiotics, pediatric dysbiosis, and disease.
Cell Host Microbe
.
2015
;
17
(
5
):
553
564
[PubMed]
23
Schulman
J
,
Braun
D
,
Lee
HC
, et al
.
Association between neonatal intensive care unit admission rates and illness acuity.
JAMA Pediatr
.
2018
;
172
(
1
):
17
23
[PubMed]
24
Harrison
W
,
Goodman
D
.
Epidemiologic trends in neonatal intensive care, 2007-2012.
JAMA Pediatr
.
2015
;
169
(
9
):
855
862
[PubMed]
25
Fisher
ES
,
Wennberg
JE
.
Health care quality, geographic variations, and the challenge of supply-sensitive care.
Perspect Biol Med
.
2003
;
46
(
1
):
69
79
[PubMed]
26
The Dartmouth Institute for Health Policy and Clinical Practice
. Supply-sensitive care. 2017. Available at: www.dartmouthatlas.org/keyissues/issue.aspx?con=2937. Accessed May 4, 2017
27
Freedman
S
.
Capacity and utilization in health care: the effect of empty beds on neonatal intensive care admission.
Am Econ J Econ Policy
.
2016
;
8
(
2
):
154
185
[PubMed]
28
Puopolo
KM
,
Mukhopadhyay
S
,
Hansen
NI
, et al;
NICHD Neonatal Research Network
.
Identification of extremely premature infants at low risk for early-onset sepsis.
Pediatrics
.
2017
;
140
(
5
):
e20170925
[PubMed]

Competing Interests

POTENTIAL CONFLICT OF INTEREST: The authors have indicated they have no potential conflicts of interest to disclose.

FINANCIAL DISCLOSURE: The authors have indicated they have no financial relationships relevant to this article to disclose.