Racial and Ethnic Differences in Emergency Department Diagnostic Imaging at US Children’s Hospitals, 2016-2019

Key Points Question Does the use of diagnostic imaging for children receiving care in US pediatric emergency departments (EDs) differ by race and ethnicity? Findings This multicenter cross-sectional study of more than 13 million pediatric ED visits to 44 children’s hospitals demonstrated that non-Hispanic Black and Hispanic patients were less likely to undergo diagnostic imaging compared with non-Hispanic White patients. Meaning In these findings, race and ethnicity appear to be independently associated with imaging decisions in the pediatric ED, highlighting the need to better understand and mitigate these disparities.


Introduction
In 2010, the American Academy of Pediatrics published a landmark report highlighting "extensive, pervasive, and persistent" disparities in pediatric health care delivery and quality. 1(p1014) An important determinant of health care quality is the appropriate use of diagnostic testing for evaluating acute illness in children. In particular, radiologic imaging for pediatric patients is commonly used in the emergency department (ED) setting, with one-third of all visits including at least 1 imaging study. 2 In addition to the many benefits, imaging also carries risks and considerations regarding resource use, including radiation exposure, 3 incidental findings leading to follow-up visits and testing, 4 increased ED length of stay, 5,6 and cost. 7 Therefore, differential use of imaging studies across racial and ethnic groups suggests that worse care is being delivered to 1 or more groups.
Studies of racial and ethnic differences in pediatric diagnostic imaging [8][9][10][11][12] have shown higher rates of selected imaging use in non-Hispanic White children compared with non-White children.
However, these studies were limited in scope, focusing on a single imaging modality for a specific condition. One study of ED visits among adults demonstrated that non-Hispanic Black patients were less likely to have radiography, computed tomography (CT), and magnetic resonance imaging (MRI) studies performed. 13 These patterns in adults may not be relevant for children, because imaging strategies, scope of presenting complaints and diagnoses, and often severity of illness differ between adults and children. [14][15][16] In our previous work, 6 we observed that non-Hispanic White children had higher odds of receiving advanced imaging compared with non-White patients. We sought to further explore this finding by evaluating whether racial and ethnic differences exist across imaging modalities and whether these differences persist across diagnoses and by insurance type.

Data Source and Study Design
This multicenter cross-sectional study of the Pediatric Health Information System (PHIS) includes administrative data from 52 tertiary care US children's hospitals. Participating hospitals are located in 27 states plus Washington, DC, representing 17 of the 20 major metropolitan areas. The Children's Hospital Association maintains the PHIS and ensures data quality and control through a joint effort with participating hospitals. We included 44 EDs in our study after excluding 8 that did not contribute complete ED data during the study period. We included all ED visits from January 1, 2016, through December 31, 2019, by patients younger than 18 years. This period was selected to enable use of the International Statistical Classification of Diseases, 10th Revision, Clinical Modification (ICD-10-CM), which was adopted in 2015 across participating sites. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline for cross-sectional studies. 17 The University of Pittsburgh institutional review board determined that the study protocol was not human subjects research and therefore was exempt from review and informed consent.

Variables and Outcome Measures
The primary outcome was the proportion of ED visits during which at least 1 diagnostic imaging test, defined as radiography, ultrasonography, CT, and MRI, was performed. These modalities were selected because they represent the most frequently performed diagnostic imaging studies in the emergency setting. 15 Diagnostic imaging in the PHIS is identified through billing codes and includes the date of imaging. However, for patients who are admitted from the ED, the data source does not distinguish between imaging performed in the ED and imaging performed as an inpatient on the same date. Therefore, and in keeping with prior work, 6 complex chronic conditions, 26 3-day revisit, 28,29 hospitalization (including intensive care unit admission), 26 visit diagnosis, 13 and year. 6 We stratified patient age into clinically meaningful categories (<1, 1-4, 5-12, and 13-17 years) and defined the visit day as weekend vs weekday and arrival time as daytime (8:00 AM to 3:59 PM), evening (4:00 to 11:59 PM), or overnight (12:00 to 7:59 AM). 30 Median neighborhood household income, presented as quartiles, was based on patient home 5-digit zip code in the PHIS and mapped to the American Community Survey 5-year data for 2014 to 2018. 31 Distance to the hospital was based on the distance between the centroids of patient home and hospital 5-digit zip codes. We defined complex chronic conditions using the system-based classification scheme by Feudtner et al, 32 which has been updated to accommodate ICD-10-CM implementation, including neonatal, technology dependence, and organ transplant categories. A visit was considered to be a 3-day revisit if an ED visit occurred within 3 prior calendar days. Given the large number of ICD-10-CM codes, we used the major diagnostic category classification system to classify visits into 1 of 26 mutually exclusive major organ system-based categories and thereby define the visit diagnosis. 33 These diagnostic categories are based on the All Patient Refined-Diagnosis Related Groups classification system, which is based on the principal discharge ICD-10-CM diagnosis for the visit 33 (eTable 1 in the Supplement). As an additional analysis, we also analyzed the top 10 principal ICD-10-CM codes responsible for the highest volume of encounters with imaging.

Statistical Analysis
We summarized data with percentages and used Rao-Scott χ 2 tests, adjusting for clustering within hospitals, to compare categorical data across race/ethnicity groups. We constructed groups of generalized linear models, including the covariates described above, with a binomial distribution and a random effect for hospital, evaluating the independent association of race/ethnicity on overall and individual imaging modalities (ie, radiography, CT, ultrasonography, and MRI). Because of the strong basis for the multicollinearity among race/ethnicity, insurance type, and median neighborhood income, 34,35 we performed a variance inflation factor analysis using a cutoff of 5. 36 For this analysis, income was estimated by race and payer, suggesting the presence of multicollinearity; therefore, we excluded income from all models. Given the large differences in insurance coverage by race and ethnicity 37 and because of the potential interaction between race and ethnicity and insurance, we replicated the modeling stratified by insurance type. The PHIS does not include data on illness severity (eg, Emergency Severity Index); in addition, non-Hispanic White race may be independently associated with lower 38 or higher 39 risk of hospitalization. Therefore, to assess the validity of our findings, we performed a separate analysis in which we limited the cohort to visits by nonhospitalized children. We used generalized linear modeling (incorporating the covariates described previously) to estimate diagnostic category-specific adjusted odds ratios (aORs) and presented those diagnostic categories that each accounted for at least 0.5% of the total ED cohort as a figure (a complete listing of data for all diagnostic categories is shown in eTable 1 in the Supplement). Finally, we used these models to calculate the adjusted proportion of visits with imaging for each race/ethnicity group. We applied the adjusted proportion of imaging in non-Hispanic White patients to the number of visits by non-Hispanic Black and Hispanic patients. We then calculated the difference in the number of visits with imaging when compared with the adjusted proportion with imaging for non-Hispanic Black and

JAMA Network Open | Pediatrics
Hispanic patients, thus establishing how many more or fewer visits would have imaging if imaging rates for visits by non-Hispanic Black and Hispanic patients were the same as those for visits by non-Hispanic White patients. Missing data were analyzed as a distinct category for relevant variables.
All hypothesis testing was 2-sided, with statistical significance defined as P < .05. We used SAS, version 9.4 (SAS Institute, Inc) for all analyses.

Characteristics of the Study Cohort
We included a total of 13 087 522 ED visits by 6

Imaging Across Diagnostic Groups Comparing Visits by Non-Hispanic White With Non-Hispanic Black Patients
Imaging was less likely to be performed during ED visits by non-Hispanic Black patients for 15

JAMA Network Open | Pediatrics
Racial and Ethnic Differences in ED Diagnostic Imaging at US Children's Hospitals

Imaging Across Diagnostic Groups Comparing Visits by Non-Hispanic White With Hispanic Patients
Imaging was less likely to be performed during visits by Hispanic patients compared with those by

Discussion
In this study of more than 13 million visits to 44 pediatric EDs, we observed that visits by non-Hispanic Black and Hispanic patients were less likely to include radiography, CT, ultrasonography, and MRI compared with those by non-Hispanic White patients. These findings were consistent across most diagnostic groups, persisted when stratified by insurance type, and were even more pronounced on analysis of only visits by nonhospitalized children. Our findings suggest that a child's race and ethnicity may be independently associated with the decision to perform imaging during ED visits.
The differential use of diagnostic imaging by race/ethnicity may reflect underuse of imaging in non-Hispanic Black and Hispanic children, or alternatively, overuse in non-Hispanic White children.
Overuse may expose these children to unnecessary risks associated with imaging. 3,4,7 Conversely, underuse may result in misdiagnoses, need for further care, and potentially worse clinical outcomes. [40][41][42] Although we were unable to discern underuse from overuse using an administrative database, it is likely that much of the imaging in White children is unnecessary. 43   higher rates of CT than non-White children, 46 even among those at the lowest risk for substantial injury. 8 There are a number of possible explanations for our findings, including a combination of parent/ guardian preferences, clinician biases, and structural factors. 47 Higher imaging rates observed in non-Hispanic White patients may, in part, be attributed to greater levels of parental anxiety with an associated increase in requests for imaging. Such a mechanism has been proposed to be a factor driving the overuse of head imaging in children at low risk of serious traumatic head injury. 8 There may also be perceived differences in the risk-balance ratio of imaging relative to radiation exposure.
A survey of adult patients in the ED reported that White patients preferred a definitive diagnostic test, such as CT, even at the expense of radiation. 48 Language barriers may also play a role. For example, non-English-speaking patients and their families may be more 49 or less 24 likely to have testing performed as part of their ED visit. Physicians' implicit racial biases are an important consideration and have been associated with patient-clinician interactions, treatment decisions, treatment adherence, and patient health outcomes. 50 These biases are exacerbated in times of stress, which is particularly relevant to ED clinicians. 51 Structural factors rooted in our health care system also likely contributed to differential imaging rates. For example, minority patients are less likely than White patients to have a medical home, 52 which may influence whether clinicians order imaging during the ED visit or defer to outpatient management, and some imaging in White children may have been driven by primary care clinician referral.
With more than 1 in every 4 ED visits in this study including an imaging study, clinicians are frequently performing diagnostic imaging. The goal, undoubtedly, assuming similar clinical presentations across racial and ethnic groups, is to enable parity in diagnostic imaging across these groups. Adherence to clinical guidelines and other objective scoring tools have the potential to reduce subjectivity, support team-based decision-making, and improve communication and structurally competent clinical care. 47,[53][54][55] Internal quality assurance evaluations to better understand physician-level practices that may be influenced by implicit bias may also narrow the disparity gap. 54,56 In addition, future work is needed to better understand hospital-level disparities in imaging delivery. Such evaluations at the hospital and clinician level are needed to enhance the quality of care delivered and health outcomes for all children.

Limitations
This study has limitations. The PHIS does not include clinical data regarding the indication for imaging, and there may be unmeasured confounders. We were unable to fully account for illness severity, given the limited clinical information contained within the PHIS (eg, Emergency Severity Index). It is possible that non-Hispanic White children had higher illness acuity, potentially accounting for higher rates of diagnostic imaging. We attempted to minimize this limitation by restricting the analysis to nonhospitalized children and observed even larger differences in imaging rates by race/ ethnicity. Race and ethnicity of some patients may have been misclassified given the varying methods of assigning race across PHIS hospitals. However, prior work evaluating race and ethnicity data in children in administrative data 57 found high accuracy in ethnicity and for White and Black race. We were unable to evaluate or control for limited English proficiency because these data are not available in the PHIS. Imaging for admitted patients may have been misclassified as having occurred as part of the inpatient stay and not the ED visit (and vice versa); notably, admitted patients were a minority of the patient population. Finally, this study was specific to US children's hospitals, and therefore, the findings are not generalizable to other EDs, care settings, or countries.

Conclusions
There are significant racial and ethnic differences in diagnostic imaging rates among children seeking care in US pediatric EDs. These differences persist across insurance groups and in analyses limited to discharged children. Further investigation is needed to better understand the factors