Invasive Bacterial Infections in Afebrile Infants Diagnosed With Acute Otitis Media

Study Design and Population

We conducted a cross-sectional study at 33 pediatric emergency departments (EDs) associated with the Pediatric Emergency Medicine Collaborative Research Committee of the AAP. The participating sites included 29 EDs in the United States, 2 EDs in Canada, and 2 EDs in Spain. All were university-affiliated teaching hospitals, and 25 (76%) were freestanding children’s hospitals. The study was approved by the Pediatric Emergency Medicine Collaborative Research Committee Steering Committee and the research ethics boards of all participating sites.

We included infants aged 90 days and younger who presented to a participating ED between January 2007 and December 2017 and were clinically diagnosed with AOM. Study period varied by site depending on available data. We excluded infants with documented temperatures of ≥38.0°C or <36.0°C in the ED or within 48 hours before ED arrival, antibiotic use other than topical antibiotics within 48 hours of ED presentation, concurrent diagnosis of mastoiditis, evidence of focal bacterial infection on ED examination, or who were transferred to the ED with diagnostic testing completed and/or antibiotic administration initiated at an outside hospital. We chose to exclude febrile infants from our study because the presence of fever in young infants, and not necessarily the diagnosis of AOM, is a primary driver for more expanded testing and/or empirical treatment of IBI.

Eligible infants were identified using two methods. The primary method was using an inclusive list of International Classification of Diseases (ICD) 9th or 10th Revision diagnosis codes for AOM (Supplemental Table 6). To assess for infants with IBI who also had AOM but for whom ICD codes for AOM were not documented, we also retrieved and reviewed the records of all infants evaluated in the ED for whom review of microbiology databases noted bacteria in the blood or cerebrospinal fluid (CSF). All diagnoses of AOM were verified through the review of clinician assessment in the medical records. Because the AAP diagnostic criteria for AOM are for children 6 months and older and may be too stringent to apply directly to this younger age group, we chose for our primary analysis to include all infants with documentation of ≥1 middle ear examination abnormality supportive of AOM diagnosis.² However, we performed a subanalysis of infants meeting simplified AAP diagnostic criteria, defined as the presence of tympanic membrane erythema, bulging tympanic membrane, or otorrhea not due to acute otitis externa.

Study Protocol

All study variables were defined a priori and described in the manual of operations. Investigators and research coordinators received standardized data abstraction training and entered their data electronically into REDCap, a secure, web-based electronic database. Collected data included demographic characteristics, medical history, presenting symptoms, ED physical examination, laboratory data, findings on assessment by otolaryngologist consultants when available, and inpatient management for hospitalized infants.

To minimize bias associated with abstracting potentially more-subjective descriptions from the medical records, we provided restrictive key words to guide the determination of ill appearance, respiratory distress, and clinical dehydration. These findings were categorized as present, not present, or unclear. To assess interrater reliability, a second abstractor at each site reviewed the medical records and performed an independent assessment of clinical appearance, respiratory status, and hydration status for a random sample of 10% of all infants, as well as for all infants with IBI or adverse outcomes.

Outcome Measures

The primary outcomes were IBIs and AOM-associated adverse events. IBIs included bacteremia and bacterial meningitis, which were defined as the growth of a pathogen in the blood culture or CSF culture, respectively (Supplemental Table 7).⁹ Any bacteria not predefined as a contaminant or pathogen, or whose categorization of pathogenicity by the treating clinician was unclear or different from the prespecified list, were reviewed by a pediatric infectious disease coinvestigator (A.C.) to determine pathogenicity. For those infants who did not have CSF obtained for culture, the presence or absence of bacterial meningitis was determined through the review of the medical records. If the infant had a follow-up visit to the index hospital within 30 days of the ED visit that did not identify bacterial meningitis, we considered this outcome as negative.

AOM-associated adverse events were defined as any substantial complications resulting from or potentially associated with AOM, including, but not limited to, the development of mastoiditis, sepsis, or death. Urinary tract infection (UTI) was not considered an AOM-associated adverse event because the causal mechanism linking AOM and UTI is unclear. However, we collected information about urine testing because it is a part of the serious bacterial infection evaluation in infants. For infants discharged from the ED, we reviewed the medical records to identify ED return visits within 72 hours. Additionally, we reviewed the medical records for any outpatient clinician encounter or hospitalization within 30 days of ED discharge. For any visit identified, we evaluated the records for evidence of IBIs or adverse events.

Our secondary outcomes were (1) ED completion of blood culture (binary), (2) ED completion of CSF culture (binary), and (3) hospitalization (binary) from the index ED visit. Each of these outcomes was stratified by age and by site to assess for variation in care.

Sample Size

Because we expected IBI to be a rare event in this sample, we determined our sample size on the basis of the prevalence of blood culture testing, a secondary outcome. Targeting evaluation of up to 20 independent variables associated with this outcome, we aimed to study at least 200 infants for whom blood cultures were obtained. Pilot testing at the lead site showed that 30% of eligible infants had blood cultures obtained; therefore, we aimed to enroll a minimum of 667 infants.

Statistical Analysis

To enhance clinical interpretability, several continuous variables were categorized. Age was categorized as 0 to 28, 29 to 56, and 57 to 90 days based on cutoffs generally used when risk-stratifying febrile infants for IBI.^10,11 The results of tests generally available during the ED stay were categorized as not performed, performed and within normal limits, and performed and abnormal. For the complete blood count, a white blood cell (WBC) count between 5 and 15 × 10³ cells/µL was considered to be normal; values outside of this range were considered abnormal.¹² Urinalysis was considered abnormal if there was any leukocyte esterase, nitrite, or >5 WBCs per high-power field, similar to previous studies.^13,14 Given the challenges of interpreting CSF WBC count if the lumbar puncture was traumatic,¹⁵ these results were not categorized.

We described the prevalence of IBI, adverse events, and diagnostic tests obtained with proportions and 95% confidence intervals [CIs]. For all tests of association regarding practice variation (eg, testing, hospitalization), we used generalized linear mixed-effects models to account for site-level clustering effects. For each outcome, we first examined the association between the outcome and a potential explanatory variable, including those from infant characteristics, clinical symptoms, and physical examination findings, using a univariable generalized linear mixed-effects model. For the subsequent multivariable models, we first assessed for potential collinearity between predictor variables using generalized variance inflation factors. Those predictors with univariable P value <.1 and no evidence of multicollinearity were included in multivariable models to assess for an association with the following outcomes as fixed effects: (1) obtaining a blood culture, (2) obtaining a CSF culture, and (3) hospitalization. Hospital sites were used as random effects.

Interrater reliability for the assessment of general appearance, respiratory distress, and dehydration was measured by using Cohen unweighted k, a 0 to 1 scale where 0 indicates poor agreement and 1 indicates perfect agreement.¹⁶ All statistical analyses were conducted using either the Statistical Program for the Social Sciences Version 26 (IBM SPSS Statistics, IBM Corporation, Armonk, NY) or R (R version 3.6.3, 2019).