Variation in the Management of Children With Deep Neck Infections

This was a retrospective cohort study, and we used data from the Pediatric Health Information System (PHIS) database, an administrative database that contains inpatient, emergency department, ambulatory surgery, and observation encounter–level data from >50 not-for-profit tertiary care pediatric hospitals in the United States. These hospitals are 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. Portions of the data submission and data quality processes for the PHIS database are managed by IBM Watson Health (Ann Arbor, MI). For the purposes of external benchmarking, participating hospitals provide discharge and encounter data, including demographics, diagnoses, and procedures. Nearly all of these hospitals also submit resource use data (eg, pharmaceuticals, imaging, and laboratory) into PHIS. Results of laboratory and diagnostic studies are not included in the database. Data are deidentified at the time of data submission, and data are subjected to a number of reliability and validity checks before being included in the database. For this study, we included data from the 39 hospitals that contributed data to PHIS for the complete study period. The study institution’s institutional review board deemed the study exempt.

We included data from children 0 to 12 years of age who were hospitalized at a PHIS-participating institution with a diagnosis of DNI between January 2010 and September 2018. We chose this age group on the basis of previous data from large-scale studies of DNIs that identified a median age of 4 to 6 years for both RPA and PPA.^2,4  We defined the diagnosis of DNI by a primary or secondary International Classification of Diseases, Ninth Revision (ICD-9) code for RPA (478.24) or PPA (478.22) or an International Classification of Diseases, 10th Revision (ICD-10) code for RPA or PPA (J39.0). In the transition from ICD-9 to ICD-10 coding in 2015, RPA and PPA were condensed into 1 code (J39.0). To increase the specificity of our case definition, we only included patients who had at least 1 antibiotic administered on the first or second hospital day. We excluded patients with a diagnosis of DNI within 30 days before the index visit as well as patients with a complex chronic condition (CCC), as previously defined.⁹  CCCs are long-term chronic conditions involving ≥1 organ system severely enough to require subspecialty care and hospitalization at a tertiary care center. Children with CCCs may be at higher risk for DNIs for a variety of reasons (eg, immunodeficiency, airway anomalies, tracheostomy tubes) and may require different management than children without CCCs. We also excluded patients transferred to a PHIS institution because they may have had imaging performed before transfer.

We defined surgical management as a relevant pharyngeal drainage procedure during the hospitalization, a version of a previously defined group of procedure codes.²  (Supplemental Table 3). We defined medical management as the lack of any of the above pharyngeal drainage procedures during the entirety of the hospitalization. We defined initial medical management as the lack of a pharyngeal drainage procedure during the first 2 days of hospitalization. We defined failed medical management as a pharyngeal drainage procedure occurring on the third hospital day or later in a patient with initial medical management. Neck radiography, ultrasounds, CT scans, and MRI scans were identified by their imaging service codes (Supplemental Table 4).

We collected demographic information, such as patient age, sex, and insurance payer, as well as clinical and administrative data, such as ICU use, charges (represented in US dollars), length of stay (LOS), and mortality. Mortality was defined as all-cause mortality during the index hospitalization. We followed patients for an additional 30 days from their index discharge date to calculate the 30-day readmission rate, both all cause and secondary to DNI. Readmissions secondary to DNI were defined as a primary diagnosis of DNI on the repeat admission. We also collected data on any repeat pharyngeal drainage procedures on those patients readmitted within 30 days.

Descriptive statistics were used to characterize the demographics of the cohort, including patient age (categorized as <6 months, 6 months to 1 year, 2–4 years, 5–9 years, 10–12 years), sex, race, ethnicity, and insurance payer. We assessed the proportion of patients undergoing imaging (radiography, ultrasound, and MRI), clinical outcomes (ICU admission, death, LOS, and rehospitalization), and hospital charges among the entire sample and stratified by management modality (medical versus surgical), comparing the 2 groups using the χ² test for categorical variables and the Wilcoxon rank test for continuous variables. We analyzed differences in the proportion of patients undergoing CT scans by patient age, sex, race, ethnicity, and payer using the χ² test. To assess for trends in surgical management over time, we estimated a logistic regression model using the surgical management indicator as the dependent variable and the study year as the independent variable and using robust SEs clustered on hospital to account for intrahospital correlation among patients.

We described hospital variation in the initial and ultimate management modality. To assess the relationship between the initial management modality and rates of failed medical management, we estimated a linear regression model with the hospital-level rate of failed medical management as the dependent variable and the rate of initial medical management as the independent variable, weighted by hospital volume of patients with DNIs.

All statistical tests were 2 tailed, and α was set at .05. The data analysis was performed by using Stata version 15.0 (Stata Corp, College Station, TX).

Over the study period, 9766 patients met inclusion criteria. After we excluded patients transferred to the study institution (n = 2262), patients with a diagnosis of RPA or PPA within 30 days before the index visit (n = 12), and patients with CCCs (n = 631), the final cohort contained 6861 patients. The median age of children in the cohort was 4.2 years (interquartile range [IQR] 2.2–6.3), and 4134 (60.3%) were male (Table 1). Only 5.8% of patients with DNIs were >9 years of age.

Overall, 5908 patients (86.1% of the cohort) underwent at least 1 neck imaging study during their hospitalization. The rates of imaging were as follows: 81.5% of patients underwent neck CT scans, 28.5% underwent neck radiography, 2.7% underwent neck ultrasounds, and 0.9% underwent neck MRI (Table 2). Hospitals had modest variation in the rate of neck CT (range 62.5%–100%) and wide variation in the rate of neck radiography (range 1.8%–71.2%). The rate of neck CT performance was higher in children managed surgically than in children managed medically (86.9% vs 77.9%, respectively; P < .001). Sixty-nine children underwent ≥2 CT scans during their hospitalization. Among children undergoing CT scans, there was no difference in the number of scans per patient between the medically managed and surgically managed group (P = .70). CT was commonly used across age groups, but its use decreased with decreasing age. CT scans were performed in 86% of children <6 months of age, 82% of children 6 to 23 months of age, 83% of children 2 to 4 years of age, 81% in children 5 to 9 years of age, and 75% of children ≥10 years of age (P = .001). The use of CT did not differ by patient sex or race. CT was used slightly more often in children of Hispanic ethnicity than in non-Hispanic children (84% vs 81%; P = .04).

The overall proportion of children managed surgically decreased from 42.0% in 2010 to 33.5% in 2018 (odds ratio 0.96; 95% confidence interval 0.93–0.99). Of surgically managed patients, the majority (82%) underwent a surgical procedure within the first 4 days of hospitalization (15% on the first day, 22% on the second day, 26% on the third day, and 19% on the fourth day).

Initial management was medical in 5008 children (73.0% of the cohort) (Table 2). Of these children, 897 (17.9%) failed medical management and underwent a surgical procedure. Over the course of the hospitalization, a total of 4111 children (60.0%) were managed medically and 2750 (40.0%) were managed surgically. Surgically managed children were younger than medically managed children (median age 3.9 vs 4.4 years; P < .001). When compared with medically managed children, surgically managed children had a higher rate of ICU admission (11.5% vs 3.2%; P < .001) and a longer median LOS (3 vs 2 days; P < .001). Median hospital charges were >$10 000 higher per visit for surgically managed children than for medically managed children ($25 241 vs $15 088; P < .001). The 30-day readmission rate did not vary between the 2 groups (12.7% vs 12.6%; P = .87). No child died, either during the initial hospitalization or during a revisit within 30 days.

Rates of overall surgical management varied widely by hospital from 16.9% to 69.6% (Fig 1). Hospitals also varied in their rates of initial medical management (range 43%–89%) and failed medical management (range 2%–40%) (Fig 2). There was a statistically significant inverse relationship between hospital-level initial medical management and failed medical management; for every 1-point increase in the hospital-level rate of initial medical management, the hospital-level rate of failed medical management decreased by 0.43 (P = .006) (Fig 3).

In this retrospective cohort study of almost 7000 children hospitalized for DNIs over a 9-year period, we demonstrated several important findings. First, the occurrence of surgical management decreased over time. Second, we documented substantial variation in management modality among US children’s hospitals. Third, children managed medically had lower health care use and charges than those managed surgically, without an increase in mortality or 30-day revisit rates. Finally, institutions with higher rates of initial medical management had lower rates of failed medical management.

The demographics and clinical characteristics of our sample are in keeping with previous studies of children with DNIs. DNIs, particularly RPAs, tend to be most common in children aged 6 months to 9 years and have a clear male predominance.^2,10–12  Although authors of previous studies have been able to characterize differences in incidence and age distribution of DNIs by specific type, our exploration was limited by the transition from ICD-9 to ICD-10 coding in 2015, which combined the diagnoses of RPA and PPA into a single code. This study is the first, to our knowledge, to evaluate patterns in imaging use in children with DNIs. We found that CT was widely used, with some variation by hospital, and was more commonly used in younger children. Importantly, we also found that medically managed children did not undergo more CT scans than surgically managed children, suggesting that the nonsurgical approach does not come at the cost of increased radiation exposure. The low number of ultrasounds performed is likely secondary to the types of DNIs explored in this study; although ultrasound is widely used to evaluate lymphadenitis and peritonsillar abscess, it is thought to have lower utility in the identification of deep space neck infections.⁵ 

There exists considerable debate about the optimal initial management of children with DNIs. Surgical incision and drainage was historically considered the gold standard of treatment but carries risks of anesthesia and neurovascular injury; furthermore, phlegmonous changes on the CT scan can be misinterpreted for an abscess, and even the presence of a rim-enhancing abscess on the CT scan has varied ability to predict the yield of pus at surgery.^13,14  Thus, over the past 20 years, DNIs have increasingly been managed with antibiotics without surgical intervention.^2,4  In a systematic review of 8 studies of children with CT-proven RPA or PPA, the authors found that among children managed medically initially, the success rate of medical management was 95%.¹  However, only one of these studies was prospective in nature, and the authors caution that the evidence lacks strength. Our study adds to the existing literature by revealing that across pediatric hospitals in a nationally representative administrative data set, only 13% of those with initial medical management eventually progressed to surgical management.

For children managed surgically in our cohort, the median LOS was 1 day longer and the median hospital charges were $10 000 higher than those for children managed medically. These results are in keeping with previous findings in which data from a decade earlier were used.^2,3  In our cohort, the rate of ICU admission was significantly higher in children managed surgically; this may be related to postsurgical monitoring or may reflect baseline differences in disease severity between the 2 groups. Notably, the 30-day revisit rate did not differ between the 2 groups, and no children in either group died. Taken together, these data suggest that a cohort of children exists that would potentially succeed with medical management and that accurately identifying that cohort could substantially reduce hospital costs and ICU use.

We excluded 2262 patients (23% of the initial cohort) who were transferred from other institutions. Our exclusion of transferred patients makes it less likely that the variation we found in rates of surgical management was solely due to referral bias (ie, community sites in some regions managing nonsurgical cases on-site and referring only cases requiring surgery), although we did not analyze practice variation at referring hospitals. One potential driver of pediatric transfers is surgical consultation at the receiving institution.¹⁵  Because the data suggest that medical management has reasonable success rates, community hospitals may consider keeping stable patients in their hospital for a trial of medical management, reserving transfer for the minority of patients who fail medical management. This is of particular importance given recently demonstrated trends of increasing regionalization of pediatric care,¹⁶  which risks increasing health care costs and decreasing overall access to care.

Interestingly, we found that hospitals with higher rates of initial medical management had lower rates of failed medical management. The reasons for this are unclear and warrant further exploration. One potential explanation is that providers at these hospitals have a higher threshold to take children with DNIs to surgery; thus, they may be more willing both to initially defer surgery and to monitor patients on intravenous antibiotics longer before declaring them failures of medical management. It may also be that centers with higher rates of initial medical management are more skilled at identifying children likely to succeed without surgical management. Our findings also reveal that children admitted later in the study period were less likely to fail medical management, suggesting that hospitals are improving their ability to manage DNI without surgery. Other factors, such as hospital patterns in antibiotic selection and catchment area (which may affect time to presentation), may also affect institutional success rates of medical management. Additional analyses are needed to further investigate these relationships.

In previous studies, it has been suggested that patients younger than 2 years of age are more likely to undergo surgical management for DNI.^6,7,17  Our data support this finding; the median age in the surgically managed group was 6 months lower than the median age in the medically managed group. The reason for increased surgical management in young children may relate to differences in airway anatomy, vaccination status, or increased illness severity; this latter possibility is supported by the observation that methicillin-resistant Staphylococcus aureus DNIs are more common in children <2 years of age.^6,7,18 

Our study had several limitations that must be considered. First, in the transition from ICD-9 to ICD-10 coding, the diagnosis codes for RPA and PPA were combined. Thus, we were unable to explore age distribution and management variation for RPA and PPA separately. Second, the use of an administrative database can lead to misclassification due to diagnostic coding errors. This is unavoidable, and we attempted to mitigate this somewhat with our inclusion criteria because children with true DNIs universally require hospitalization and antibiotics. Third, the PHIS database does not provide results of laboratory studies or imaging; thus, we were unable to explore the influence of laboratory and radiographic findings on management modality. This is particularly significant because multiple previous studies have suggested that a larger abscess size on the CT scan is a significant factor predicting the need for surgical management.^6,7,17  Fourth, the lack of microbiologic data precludes an evaluation of the causative pathogens of DNIs. Studies evaluating the relationship between causative pathogen and failure of medical management are crucial to better guide clinical care, particularly when methicillin-resistant Staphylococcus aureus is concerned. In future research, authors should also explore whether children managed medically receive different antibiotics or have a longer duration of antibiotic therapy than children managed surgically. In addition, individual patient vaccination status, which is not available in PHIS, may have impacted the severity of illness and the underlying microbiology. Finally, our study reflects the management of children at tertiary care pediatric hospitals in the United States. The management of children with DNIs who are cared for in nonpediatric hospitals or international settings may differ.

In this study, we found that although the overall rate of surgical management of DNIs is decreasing over time, there remains substantial variation in management modality among US children’s hospitals. Surgical management is associated with demonstrably higher resource use and cost, with no difference in 30-day readmissions. Large-scale patient-level prospective studies are needed to identify children with DNIs who are likely to succeed with medical management in order to standardize and improve care.