Hospital-onset bacteremia in the neonatal intensive care unit: strategies for risk adjustment

Abstract

Objective: To quantify the impact of patient- and unit-level risk adjustment on infant hospital-onset bacteremia (HOB) standardized infection ratio (SIR) ranking.

Design: A retrospective, multicenter cohort study.

Setting and participants: Infants admitted to 284 neonatal intensive care units (NICUs) in the United States between 2016 and 2021.

Methods: Expected HOB rates and SIRs were calculated using four adjustment strategies: birthweight (model 1), birthweight and postnatal age (model 2), birthweight and NICU complexity (model 3), and birthweight, postnatal age, and NICU complexity (model 4). Sites were ranked according to the unadjusted HOB rate, and these rankings were compared to rankings based on the four adjusted SIR models.

Results: Compared to unadjusted HOB rate ranking (smallest to largest), the number and proportion of NICUs that left the fourth quartile (worst-performing) following adjustments were as follows: adjusted for birthweight (16, 22.5%), birthweight and postnatal age (19, 26.8%), birthweight and NICU complexity (22, 31.0%), birthweight, postnatal age and NICU complexity (23, 32.4%). Comparing NICUs that moved into the better-performing quartiles after birthweight adjustment to those that remained in the better-performing quartiles regardless of adjustment, the median percentage of low birthweight infants was 17.1% (Interquartile Range (IQR): 15.8, 19.2) vs 8.7% (IQR: 4.8, 12.6); and the median percentage of infants who died was 2.2% (IQR: 1.8, 3.1) vs 0.5% (IQR: 0.01, 12.0), respectively.

Conclusion: Adjusting for patient and unit-level complexity moved one-third of NICUs in the worst-performing quartile into a better-performing quartile. Risk adjustment may allow for a more accurate comparison across units with varying levels of patient acuity and complexity.

Figure 1.

Scatter Plots of Neonatal Intensive Care Unit-level Characteristics with Site Hospital-Onset Bacteremia Rates. Correlation coefficients are shown as r_s.

Figure 2.

Alluvial plot of neonatal intensive care unit (NICU) rankings based upon the unadjusted HOB rate and standardized infection ratios (SIR) calculated from four adjusted models: birthweight (model 1), birthweight and postnatal age (model 2), birthweight and NICU complexity (model 3), and all variables (model 4). Based upon unadjusted HOB rate, the sites in the fourth quartile (worst-performing) are shown in dark gray and first-third quartiles (better-performing) are shown in light gray. Forty-four sites remained in the fourth quartile and 185 sites remained in the first-third quartiles regardless of adjustment. Across all adjustment strategies, 55 sites experienced a change into or out of the fourth quartile. The plot is truncated to show the 55 sites that experienced a change in performance quartiles and is not to scale.

Figure 3.

Scatterplots, with Spearman correlation coefficients, of HOB Standardized Infection Ratio (SIR) rank (ordered smallest to largest) derived from adjusted SIR model 1 compared to adjusted SIR model 2–4 (Panels A through C) for 284 neonatal intensive care units (NICUs) in the analysis. Risk adjustments include: birthweight (model 1), birthweight and postnatal age (model 2), birthweight and NICU complexity (model 3), and birthweight, postnatal age, and NICU complexity (model 4). Panel D displays adjusted SIR model 4 based on all HOB to the corresponding SIR rank using only non-commensal and treated commensal HOB events based on 277 NICUs that provided antibiotic data.