Trends in ED Resource Use for Infants 0 to 60 Days Evaluated for Serious Bacterial Infection

We performed a multicenter cross-sectional study using data from the Pediatric Health Information System (PHIS), an administrative database that contains inpatient, ED, ambulatory surgery, and observation encounter-level data from 52 tertiary care pediatric hospitals in the United States affiliated with the Children’s Hospital Association (Lenexa, KS). Data quality and reliability are ensured through a joint effort between the Children’s Hospital Association and participating hospitals. Data are deidentified at the time of data submission and subjected to reliability and validity checks before being included in the database. Data available in the PHIS include demographics, diagnostic codes, procedural codes, and billing codes for pharmacy, imaging, and laboratory testing; clinical data, including laboratory test results, vital signs, and clinician notes, are not available. For this study, we included data from 27 hospitals after excluding 3 hospitals with data quality issues and 22 hospitals with incomplete data. The study was approved by the institutional review board and the administrators of the PHIS database. All statistical analyses were performed by using Stata/SE version 14.1 (Stata Corp, College Station, TX).

ED encounters for infants aged 0 to 60 days over the 10-year period between January 1, 2010, and December 31, 2019, were eligible for inclusion in this study. Encounters were included if they had both a blood culture and a urine culture sent on the day of the index ED visit. Index ED visits were defined as any ED visit meeting inclusion criteria during the study period if the infant did not have a previous visit within the preceding 3 days.

To restrict the sample to otherwise healthy, well-appearing infants, index encounters with the following were excluded: patients with complex chronic conditions,¹⁵  patients admitted to the ICU, patients who required respiratory support the day of the index ED visit (oxygen, noninvasive positive pressure ventilation, high-flow ventilation, or mechanical ventilation), patients who received vasoactive support on the day of the index ED visit (parenteral dopamine, epinephrine, or norepinephrine), and patients who died during the index encounter.

LP was defined as either performance of an LP by using International Classification of Diseases procedure codes or CSF testing, including CSF cell count, culture, or fluid analysis, on the day of the index encounter. Antibiotic administration was defined as billing codes for ≥1 of the following parenteral antibiotics during the index encounter: ampicillin, gentamicin, ceftriaxone, cefotaxime, cefepime, or vancomycin. Procalcitonin testing was defined as performance of testing on the day of the index ED visit. Revisits were defined as any ED visit occurring within the subsequent 3 days after an index ED visit. Diagnosis of SBI (which was only analyzed for visits in 2017–2019) was defined by International Classification of Diseases, 10th Revision (ICD-10) codes assigned to the ED revisit as listed in Supplemental Table 4.

We summarized the demographic and clinical characteristics of index encounters using frequencies and percentages or medians and interquartile ranges (IQRs) for categorical and continuous variables, respectively. We stratified encounters by procalcitonin testing, both for the entire study period and restricted to the years 2017–2019. All tests were 2-tailed, and α was set at .05.

For our primary analysis describing trends in ED resource use, we examined linear temporal trends for the years 2010–2019 using a set of logistic regression models with LP, antibiotic administration, hospitalization, and procalcitonin testing as the dependent variables and time (measured in years) as the independent variable. These models were used to test the null hypothesis of no change in the odds of the outcomes over time (ie, odds ratio = 1). Effect estimates were measured as the percentage change in the odds for each 1-unit increase in time (yearly or quarterly, as applicable). We also examined quarterly trends in procalcitonin testing and LP by hospital for the years 2017–2019.

For our secondary analysis describing the association between procalcitonin testing and resource use under more current practice, we restricted the analysis to the 3 most recent years (2017–2019). We estimated a set of multivariable logistic regression models with encounter-level LP, antibiotic administration, and hospitalization as the dependent variables and hospital-level procalcitonin testing, defined as the proportion of encounters within a given hospital that included a procalcitonin test, as the independent variable. In addition, on the basis of previous literature and our clinical experience, we selected the following variables a priori for inclusion in the model as measures that may be associated with variability in resource use in the ED setting: calendar year of encounter, patient age in days, insurance status, race, and ethnicity.¹⁶  We used a robust variance estimator to accommodate the correlation resulting from the clustering of patients within hospitals. To assess for possible outliers in terms of resource use, we also performed a sensitivity analysis in which we excluded any hospitals with LP rates <25% on the basis of rates of LP observed in the literature.^17,18  We conducted analyses separately by age subgroups: 0 to 28 days and 29 to 60 days.

We also tested the association between ED revisits and procalcitonin testing for the years 2017–2019. We estimated a multivariable logistic regression with ED revisit within 3 days as the dependent variable and procalcitonin testing and the demographic factors listed above as the independent variables. To evaluate for potential delays in care of adverse outcomes, we repeated this model 4 times using ED revisit with each of the following as the dependent variable, in turn: LP, antibiotic administration, hospitalization, and SBI.

Eligibility screening was performed on 117 167 encounters (Supplemental Fig 3). After applying our inclusion and exclusion criteria, 107 022 encounters from years 2010 to 2019 were included in this study, of which 106 547 (99.6%) were index visits and 475 (0.4%) were revisits. Demographic characteristics for index encounters from years 2010 to 2019 and years 2017 to 2019 are shown in Table 1. Overall, demographic characteristics were similar between infants who had procalcitonin testing performed and those who did not for both study periods. The majority of procalcitonin testing (91.1%) was performed during the years 2017–2019. The median age was similar for infants who underwent procalcitonin testing and those who did not (median [IQR]: 34 days [19–47] and 33 days [18–46], respectively) across both study periods.

For the years 2010–2019, age-stratified trends in LP, antibiotic administration, hospitalization, and procalcitonin testing are shown in Fig 1. Adjusted annual odds ratios (aORs) for LP, antibiotic administration, and hospitalization, stratified by age, are shown in Table 2. For LP, antibiotic administration, and hospitalization, annual adjusted odds decreased more for infants aged 29 to 60 days compared with those aged 0 to 28 days (Table 2).

Data for quarterly procalcitonin testing and LP rates for 2017–2019 at each study hospital are displayed in Fig 2. LP rates by hospital ranged from 7% to 84%. No consistent relationship between rates of procalcitonin testing and LP was observed. Approximately one-quarter of the hospitals (7 of 27; 25.9%) had quarterly procalcitonin testing rates of <5% for the entirety of 2017–2019. Only 3 hospitals had procalcitonin testing rates of at least 50% for the entirety of 2017–2019.

Age-stratified results of the multivariable regression models for hospital-level procalcitonin testing and (1) LP, (2) antibiotic administration, and (3) hospitalization for 2017–2019 are shown in Table 3. There was no association between hospital-level procalcitonin testing and LP, antibiotic administration, or hospitalization for either subgroup (Table 3). In our sensitivity analysis excluding the 3 hospitals with LP rates <25%, there remained no association between hospital-level procalcitonin testing and LP (aOR 0.78 [95% confidence interval (CI) 0.40–1.53] for age 0–28 days; aOR 1.44 [95% CI 0.98–2.12] for age 29–60 days).

Encounters that included procalcitonin testing were not more likely to result in an ED revisit within 3 days compared with encounters that did not include procalcitonin testing (aOR 1.25 [95% CI 0.81–1.92]). Similarly, compared with encounters that did not include procalcitonin testing, encounters that included procalcitonin testing were not more likely to result in an ED revisit with LP (aOR 1.33 [95% CI 0.73–2.42]), antibiotic administration (aOR 1.15 [95% CI 0.58–2.29]), hospitalization (aOR 1.29 [95% CI 0.80–2.08]), or diagnosis of SBI (aOR 0.88 [95% CI 0.50–1.58]). Results for each outcome remained similar after adjustment for LP at the index visit.

In this large cross-sectional multicenter study of resource use for infants aged 0 to 60 days undergoing ED evaluation for SBI, we found that rates of LP, antibiotic administration, and hospitalization decreased more for encounters of infants aged 29 to 60 days from 2010 to 2019 compared with those of infants aged 0 to 28 days. We found that procalcitonin testing has recently increased. Hospital-level procalcitonin testing was not associated with changes in LP, antibiotic administration, or hospitalization at the index visit, nor was it associated with an increase in revisit rate or missed diagnosis of SBI.

Previous studies have revealed that there is wide variation in the evaluation and management of febrile infants. In a study of >2000 infants ≤28 days evaluated for fever in the ED, only 73% received blood, urine, and CSF cultures, whereas 79% received appropriate antibiotics and 84% were admitted.¹⁹  A study of febrile infants aged 7 to 90 days found similar variability in resource use and also found that fewer infants aged 29 to 60 days received blood, urine, and CSF cultures, as well as antibiotics and hospitalization, compared with neonates aged 7 to 28 days.¹⁸  We found similar variability in resource use for these infants. Additionally, we observed that this variability has changed over time, particularly among infants evaluated for SBI with blood and urine cultures. This observed temporal decrease in resource use provides important context to the evaluation of febrile infants as more institutions move to implement procalcitonin testing into the evaluation for this population.

Procalcitonin testing in other disease states is associated with changes in patterns of clinical care. For example, procalcitonin testing was associated with decreased rates of antibiotic prescribing for hospitalized children with community-acquired pneumonia and decreased length of antibiotic administration for early-onset sepsis in neonates in the ICU.^20–22  Although several studies have revealed the superior test performance of procalcitonin in the ED evaluation for SBI in young infants, the association between procalcitonin testing and resource use for this population is not well described.^{10,11,23–26}  In 2019, Kuppermann et al¹⁴  published a clinical prediction rule incorporating procalcitonin testing into the evaluation of febrile infants <60 days. Although procalcitonin testing is only applied for risk stratification in a subset of infants, the authors reported that approximately two-thirds of low-risk infants aged 29 to 60 days who received LPs could have safely avoided this procedure had the prediction rule been applied.¹⁴  Although these data represent a promising potential for decreased resource use for these infants, a temporal lag often exists between high-quality knowledge generation and knowledge translation. In a single-center prospective study of ED patients aged 1 to 36 months undergoing evaluation for fever without a source, physician knowledge of procalcitonin testing results was not associated with any difference in chest radiograph or LP rates, antibiotic administration, or hospitalization.²⁷  The authors explained that the main limitation of their study was that physicians had not used procalcitonin testing results in their decision-making before, and thus their results may be due to a delay in knowledge translation. We posit that a similar phenomenon is influencing overall patterns of resource use even in centers with procalcitonin testing available. Additionally, we did observe a decrease in rates of LP in some individual hospitals after procalcitonin testing increased. In addition to whether a procalcitonin test was obtained, interpretation of procalcitonin levels varies in previous studies, and therefore its value may not be fully realized without additional studies on implementation and outcomes.^13,14  Given the potential impact of procalcitonin testing on resource use and given our findings, future research on optimal implementation is needed.

Our study has several important limitations. First, in this study, using a large administrative database, we attempted to approximate a population of otherwise healthy infants undergoing ED evaluation for SBI using our inclusion and exclusion criteria. We included patients who had blood and urine cultures sent on the first day of their visit. As a result, we may have missed infants who presented close to midnight and had cultures sent on different calendar days. Furthermore, procalcitonin testing has been validated for febrile infants but not in other populations of infants, and our sample may have included some infants who did not have fever but instead were hypothermic or had another indication for an SBI evaluation. In addition, although we believe that procalcitonin testing was consistently captured in PHIS because laboratory testing in general is reliably captured by PHIS, we do not have matching clinical data to verify. Finally, although the use of procalcitonin testing has increased over time, its use in this population is still relatively low. Because the rates of procalcitonin testing remain relatively low, it is difficult to fully evaluate any significant differences in resource use, even if such differences exist. The increase in procalcitonin testing in our study population is recent, with rates only beginning to approach 30% in 2019. There is likely to be a temporal lag between the increase in procalcitonin testing and changes in resource use, which require changes in clinician and institutional practice. Subsequent studies when procalcitonin testing is more widespread, may reveal further changes in patterns of resource use that were not evident in this study.

We found that rates of LP and antibiotic administration decreased significantly from 2010 to 2019 for infants aged 29 to 60 days undergoing ED evaluation for SBI, whereas procalcitonin testing remained relatively low but has increased recently. Despite this recent increase, there was no association between hospital-level procalcitonin testing and LP, antibiotics administration, or admission. As resource use in young infants being evaluated for SBI continues to evolve and as procalcitonin testing becomes more widespread and well established, optimal implementation strategies to improve clinical outcomes require further study.