Low-Value Diagnostic Imaging in Children with Medicaid

We obtained data from Pennsylvania Medicaid, the fifth largest Medicaid program in the country, and the single largest health insurer in the state.( 12 ) We conducted a retrospective longitudinal cohort study of children enrolled in Medicaid using inpatient, outpatient, and professional claims, as well as enrollment data, for the period spanning January 1, 2014 to December 31, 2016. Data were used for this analysis through an intergovernmental agreement between the University of Pittsburgh and the Pennsylvania Department of Human Services. The University of Pittsburgh Institutional Review Board determined the study protocol was not human subjects research and, therefore, was exempt from review. We included patients <18 years-old at the start of the calendar year and enrolled in Pennsylvania Medicaid during 2014–2016 who were not dually enrolled (in Medicaid and Medicare) at any point during the study period. We excluded those with complex chronic conditions using the pediatric complex chronic conditions classification system( 13 ) for diagnoses included in claims during the study period. These conditions are defined as involving several organ systems or one organ system severely enough to require specialty care and likely periods of hospitalization at a tertiary care center. ( 14 ) We focused on patients with the following 5 pediatric conditions for which imaging is not routinely recommended as part of the Choosing Wisely campaign: bronchiolitis (chest x-ray),( 15 ) minor head trauma (head CT),( 9 , 16 ) simple febrile seizure (head CT or magnetic resonance imaging (MRI)),( 9 ) uncomplicated headache (head CT or MRI),( 17 ) and abdominal pain (abdominal CT without prior abdominal ultrasound (US)).( 9 , 17 , 18 ) To capture patients with these conditions, we included claims with the relevant principal International Classification of Diseases, Clinical Modification, 9 th (ICD9) and 10 th editions (ICD10) diagnosis codes. Inclusion and exclusion codes were specific to the conditions and based on prior work as well as relevant literature. ( 7 , 19 , 20 ) ( Table 1; online ). We defined episodes of care as occurring in either an emergency department (ED), inpatient hospital unit, outpatient clinic, including primary care and subspecialty visits, or “other” setting. We defined ED encounters among outpatient and professional claims using revenue codes, procedure codes, and place of service, with associated claims identified by matching claim numbers, billing provider, and date. Encounters were considered inpatient if they occurred during an inpatient hospital stay, inclusive of admission and discharge dates on an inpatient facility claim, were not under “observation” CPT codes, or were identified from professional claims and labeled as having occurred in an inpatient hospital. Outpatient clinic encounters, which included primary care visits and subspecialty visits, were defined based on professional and outpatient claims that did not occur during an inpatient stay and did not have ED procedure or revenue codes. Additionally, all professional claims included a place of service indicating the visit took place in an office or clinic setting. Encounters with a code for observation status were included in either the ED or outpatient population, and represented <1% of our samples encounters. Prior work has demonstrated variations in pediatric imaging across healthcare settings,( 20 – 24 ) therefore, in order to explore such variations with low-value imaging services, we further stratified inpatient and ED encounters based on pediatric status of the treating facility. Specifically, we linked Medicaid claims to data from the American Hospital Association (AHA) and Healthcare Cost Report Information System (HCRIS). Children’s hospitals were defined based on their classification as a pediatric facility in either HCRIS or AHA data.( 25 ) We identified pediatric EDs not in a children’s hospital, as well as encounters in pediatric units not housed within children’s hospitals based on classification in AHA data. In some cases, we could not classify encounters due to concerns about appropriate matching between Medicaid claims and HCRIS/AHA data (8.8% of ED and 3.0% of inpatient encounters,) and these remaining encounters were labeled “unclassified.” Our primary outcome was count of low-value imaging services included as part of an encounter for the 5 specified conditions. Imaging was identified using procedure and revenue codes on the day of, one day prior, or one day after the date of service using current procedural terminology (CPT) and revenue codes ( Table 1; online ) relevant to the conditions of interest (e.g. chest XR in bronchiolitis, CT and MRI in febrile seizure). We included days adjacent to the billing date to account for a lag either in performance of an imaging study or in the billing date assigned. For abdominal pain encounters, we defined a low-value CT as that without an US performed within 7 days prior to the CT. We selected 7 days as a conservative time period to reduce misclassification of a CT as low-value and performed a sensitivity analysis varying the time period for US to evaluate the impact of this definition. In addition, we applied a more liberal definition of low-value CT and evaluated the rate of CT regardless of whether US was performed in the 7 days prior. We calculated total costs based on the paid amount for the imaging service, identified by the professional and outpatient claims with relevant CPT and revenue codes. Because the data did not allow for isolation of facility costs from imaging, specifically for admitted patients, we present expenditures only for ED and outpatient encounters. We categorized age into 5 clinically meaningful categories. Race and ethnicity are self-reported in the Medicaid enrollment data. We present results by non-Hispanic White, non-Hispanic Black, Hispanic, and all other race/ethnicity groups. Urbanicity was determined using patient ZIP code and rural urban commuting area codes (in which any metropolitan category was considered “urban.”( 26 )) Patients were also assigned to one of five managed care contracting regions based on county. We included out-of-state encounters for patients who received care outside of Pennsylvania because Pennsylvania Medicaid served as the primary payer for the encounter. We grouped reasons for Medicaid eligibility into disabilities and chronic medical diagnosis, children and families, and “other” (<1%). Chronic Illness and Disability Payment System (CDPS) ( 27 ) is a diagnostic classification system that Medicaid programs use for risk adjustment and health-based capitated payments and includes less severe diagnoses (e.g. migraine, cellulitis, endometriosis)( 28 ) than those identified as part of the complex chronic condition classification.( 13 ) For this variable we counted the number of diagnosis groups applicable to each enrollee based on diagnosis codes accumulated during the calendar year. We also categorized the managed care organizations (MCO) that paid for the low-value services. The state is divided into 5 regions, and >99% of beneficiaries are enrolled in one of the state’s managed care organizations (MCO), with the remainder in fee-for-service Medicaid. We performed a descriptive analysis of the rate of low-value imaging services across each of the 5 conditions as well as summarized across all conditions. The denominator for each condition was the number of encounters by eligible patients with the condition. For each low-value imaging study, we identified the number of low-value services associated with each diagnosis and calculated rates per 1000 patient encounters with 95% confidence intervals. We performed a similar analysis of the rate of low-value imaging services by healthcare setting. For the analysis of costs, we summed paid amounts on professional and outpatient claims for low-value services by condition and present the total amount paid. We estimated rates of services among categorical variables using a simple Poisson regression with encounter as the unit of analysis, count of low-value imaging as the dependent variable, and patient characteristics described above as the independent variables. To evaluate the independent association of each of the demographics described above with low-value imaging, we performed a multivariable Poisson regression. Analyses were performed using SAS version 9.4 and STATA version 15.1. We considered p-values less than 0.05 statistically significant.

There were 1,510,278 children enrolled in Pennsylvania Medicaid from 2014–2016. After excluding those dually enrolled or partially eligible as well as those with complex chronic conditions, there were 1,305,981 children eligible, representing 3,066,096 person-years, for analysis. Identifying those with qualifying conditions and applying relevant exclusion criteria resulted in a total of 640,450 encounters among 362,991 patients across the 5 conditions Figure 1; online . Of the 640,450 encounters for the 5 conditions, there were 36,262 low-value imaging services (56.6 [95%CI: 56.0, 57.2] per 1000 encounters) delivered to 31,455 patients ( Figure 1; online ). Compared to those without low-value imaging performed, a higher proportion of patients who received low-value imaging were White (58.5% vs 50.1%), lived in rural communities (20.6% vs 13.6%), and had ≥1 CDPS flag (66.6% vs 36.0%) ( Table 2 ). The rates of low-value imaging services ranged from 19.5 abdominal CTs per 1000 encounters for abdominal pain, to 246.0 chest x-rays per 1000 encounters for bronchiolitis ( Table 3 ). Varying the number of days US was performed prior to CT as well as defining a low-value CT as with or without a prior US did not alter the rates of low-value CT for encounters for abdominal pain, as there were no US studies performed between 1 and 7 days prior to CT. Rates of low-value abdominal CT increased with patient age, and children <6 years had higher rates of neuroimaging for head trauma and headaches compared to older children ( Table 3 ). There were higher rates of advanced imaging (i.e. CT, MRI) among White children relative to Black children, and compared to children in urban communities, those living in rural areas had higher rates of low-value imaging across all conditions. Adjusting for relevant confounders, Black and Hispanic children were more likely to receive a chest x-ray for bronchiolitis ( Table 4 ). Black children were also more likely to have a head CT for minor head trauma, but less likely to receive an abdominal CT for abdominal pain. Children in urban areas were less likely to have abdominal CT imaging for abdominal pain, and neuroimaging for headache diagnoses. The likelihood of low-value imaging varied by region and patients in several MCOs were less likely to receive low-value imaging than those not in MCOs, particularly for advanced imaging services (i.e. CT and MRI). Among the 640,450 encounters, 59.1% were in the outpatient setting, 34.9% were in the ED, 2.2% were in the inpatient setting, and 3.8% were in “other” settings ( Table 5; online ). Low-value imaging studies occurred in 973 EDs, 458 inpatient units, 826 offices/clinics, and 0 “other” settings. EDs performed between 53.4% (neuroimaging for headache) and 87.9% (CT for minor head trauma) of low-value imaging studies; 9.6% (CT for minor head trauma) to 43.7% (neuroimaging for headache) were performed in the outpatient setting; and, 2.5% (CT for minor head trauma) to 20.7% (neuroimaging for febrile seizure) were performed during inpatient encounters ( Figure 2 ). Of the low-value imaging studies performed in the ED, non-pediatric EDs were responsible for the largest proportions for all conditions: between 55.9% (neuroimaging for febrile seizure) and 77.2% (abdominal CT without US for abdominal pain). Remaining low-value imaging in the ED was performed with similar frequencies in pediatric EDs within children’s and non-children’s hospitals ( Figure 2 ). Rates of low-value imaging among patients with the condition in each setting are shown in Table 5; online . Among patients with bronchiolitis, EDs had the highest rates of imaging. While the inpatient setting had the highest rates for other conditions, the numbers of encounters for these conditions were much lower than those for the ED and outpatient setting. The cost of low-value imaging in the ED was $4,512,613.08, and $2,785,652.55 was spent in the outpatient setting.

Our study indicates that more than 31,000 low-value imaging services were performed for 5 pediatric conditions studied over a 3-year period, resulting in one low-value imaging study for every 18 eligible encounters. The most striking examples of low-value imaging were seen with encounters for bronchiolitis, minor head trauma, and headache, with one in every 4, 6, and 7 encounters, respectively, including a low- value imaging study. White children and those in rural communities received a higher proportion of these studies. While only a third of encounters for these 5 conditions were in the ED, the ED was responsible for nearly three quarters of the low-value imaging services, with the largest proportion in non-pediatric EDs. These imaging services were responsible for $7.3 million in ED and outpatient costs. Most of the literature regarding low-value care to date has focused on adults patients.( 29 ) Given the disproportionate financial cost of imaging studies on the overall burden of pediatric low-value care,( 7 ) and the inverse relationship between radiation risk and age of exposure,( 30 ) it is important to understand the epidemiology of these low-value tests. Studies by Chua, et al,( 7 , 10 ) which included two of the imaging tests in our study, have improved our understanding of low-value services in children. Our work builds upon this by focusing on an expanded list of low-value imaging measures as well as exploring the settings responsible for these services. Our finding that White children are independently more likely to receive abdominal CT imaging for abdominal pain compared to Black children is consistent with prior work( 20 , 31 ) and may also reflect other findings that Black children are more likely to have undertreated abdominal pain. ( 32 ) Although others have found that White children are more likely to receive low-value x-rays and head CTs for bronchiolitis and head trauma, respectively, ( 33 , 34 ) our findings suggest that Black children may be more likely to have these studies performed. In the case of head CTs, this finding may relate to data indicating that Black children are more likely to be evaluated for non-accidental trauma. ( 35 ) In addition, prior studies did not focus on Medicaid patients, exclusively, and our findings may reflect differential access to care and perceived access to follow-up across racial groups( 36 , 37 ) which can influence physician decisions. Nevertheless, the association of race/ethnicity with low-value imaging warrants further study to better understand the contributions of patient-level factors and potential physician implicit biases.( 38 ) We found that the ED was responsible for the majority of the observed low-value imaging. In addition, most of this imaging occurring in non-pediatric EDs. While this is consistent with the fact that more than 80% of pediatric visits to EDs in the U.S. occur in EDs that do not specialize in pediatric care,( 39 ) these findings clarify the need for interventions to reduce low-value imaging to focus on this setting. Although pediatric EDs were responsible for the minority of low-value imaging services, these EDs were not absolved from contributing to the overall burden, which has been highlighted in other work.( 40 ) There are several potential explanations why the ED was a significant source of low-value imaging. Pediatric Medicaid patients in our study, similar to a finding in a study of 6 geographically-dispersed states,( 41 ) may disproportionately rely on ED care compared with commercially insured patients. This could result in more testing being done in the ED. Obtaining ED imaging such as radiographs and CTs is more convenient than in other settings, with many EDs having dedicated x-ray and CT suites; for patients with acute concerns (e.g. bronchiolitis, head injury) outpatient offices may refer patients to the ED for timely evaluation and management and defer imaging decisions to the ED as well. As many as 18% of low-value x-rays for bronchiolitis, and 20% of neuroimaging for febrile seizures were performed in the inpatient setting. Although definitions of low-value care are often limited to outpatient and ED services( 7 , 10 , 40 ) it is important for inpatient care to be scrutinized as well. Interventions have been studied to specifically reduce inpatient low-value imaging, and our data suggest further work is needed which targets this population.( 42 , 43 ) We also observed variability in low-value imaging services across MCOs, suggesting that some MCOs may benefit from implementing policies and working with providers to reduce these imaging rates. There are several reasons why clinicians order low-value testing. Campaigns and outreach efforts, ( 44 , 45 ) have all brought to light the low-value services clinicians incorporate into patient care. ( 8 , 46 ) Although there has been widespread advertising and promotion of the underpinnings of these campaigns by professional societies and media outlets, we may not have (yet) seen the intended consequences of this effort.( 47 ) Survey studies reveal a desire for more information to reduce uncertainty, “just to be safe,” malpractice concerns over errors of omission, patients’/caregivers’ insistence, and a desire to keep patients/caregiver happy, as important drivers of low-value care.( 48 , 49 ) In particular, there has been recent emphasis on patient satisfaction( 50 ) and prior work highlights the influence of parental anxiety/request as a factor in low-value imaging.( 34 ) Another consideration is resources available to clinicians (e.g. local pediatric-specific policies, guidelines, and practices which may limit low-value care), particularly given much of the low-value imaging we observed was in non-pediatric EDs which may not have these available. It will be important in future work to evaluate temporal trends in low-value studies to evaluate if improvements have been made over time. Similarly, analyses of the short and long-term consequences of low-value imaging will be important for future studies. Mitigating the burden of low-value services, including unnecessary imaging studies, likely requires a multipronged approach.( 51 , 52 ) Reports of effective interventions include guideline implementation, education, and provider feedback, which show promise in reducing pediatric low-value imaging studies in the inpatient and ED settings in pediatric( 43 , 53 – 57 ) and community hospitals.( 58 ) Strategies aimed at closing the gap in access to pediatric-subspecialty care may also reduce low-value imaging testing. Telemedicine, in particular, has been shown to reduce unnecessary ED visits, particularly in remote areas,( 59 ) which, in our study, demonstrated a disproportionately higher rate of low-value imaging (compared with rates in other settings). If telehealth can safely keep a portion of children with these diagnoses at home, these patients also become substantially less likely to receive imaging during the episode of illness. Additionally, telehealth serves may allow on-demand pediatric ED consultation to non-pediatric EDs and similarly, outpatient generalist providers may use telehealth to gain expert decision support from a subspecialist before ordering low-value imaging for patients with headaches and febrile seizures, for example. Until recently, the uptake of telehealth has been slow.( 60 ) However, given the recent and unexpected need for telehealth services in the context of the COVID-19 pandemic, many of the prior financial and organizational barriers have been removed, at least temporarily.( 61 ) As telehealth use likely enters into a new steady state in the future, it will be important to monitor the impact on low-value care. Finally, we may see more impactful changes in low-value services with financial incentives and penalties ( 62 , 63 ) for providers and health care delivery organizations. In the current healthcare climate, providers lack sufficient restrictions on low-value ordering practices as well as incentives for high value care. Medicaid programs may consider emulating initiatives such as Medicare’s Appropriate Use Criteria Program and Quality Payment Program, both of which financially reward high-value care in imaging services. ( 64 , 65 ) There are important limitations to our work. First, our data did not contain presenting complaints and we relied on discharge diagnoses (which can be influenced by diagnostic imaging) to identify conditions of interest. For each condition, we excluded those associated diagnoses that may indicate appropriate imaging. However, there may be low-value imaging in this excluded population resulting in an underestimate of our rates. Conversely, without clinical data, it is possible we overestimated these low-value imaging rates. Patients are not necessarily continuously enrolled in Medicaid since birth, and therefore, not all complex chronic conditions may be present in claims during the study period. However, we believe this misclassification would be minimal, as these conditions would likely be those for which patients did not receive frequent care. There was a higher proportion of patients with CDPS flags in the low-value imaging group, however, these identified diagnoses include those not captured by our condition-specific exclusion criteria and, therefore, should not significantly impact care for that condition. We were unable to classify some EDs and inpatient units based on pediatric status, however, these represented only a small proportion of locations delivering low-value imaging services. In addition, we were unable to stratify outpatient visits by primary care vs subspecialty care. Although responsible for the minority of low-value imaging in our study, it should be noted that defining low-value imaging during inpatient encounters may be more challenging and deserves further study. Finally, our study focused on those patients enrolled in Medicaid, and therefore, our results may not generalize to all pediatric patient populations. However, other studies comparing rates of low-value services between Medicaid and commercially insured populations found small and inconsistent differences between payer type.( 10 , 66 ) Additionally, we focused on a single state and given that there is variation in low-value care delivery across states,( 10 ) our findings may not be comparable to other states. Our findings provide policy-relevant insight into the prevalence of low-value imaging across several pediatric conditions and settings of care. These data can inform targeted interventions to steer providers towards best practices and minimize the burden of many low-value imaging studies on children and Medicaid programs.

Low-value services are those that do not result in a net benefit for patients.( 1 ) Diagnostic imaging studies provide clinical information rapidly and can reduce diagnostic uncertainty. However, imaging also carries the risks of radiation( 2 , 3 ) in the cases of plain radiography (x-ray) and computed tomography (CT), and false positive and incidental findings and their downstream effects, including additional medical visits for families and testing.( 4 – 6 ) In addition, low-value imaging tests are responsible for significant and potentially avoidable costs to payers and patients. A 2016 analysis of commercially insured pediatric patients estimated that although imaging tests are responsible for only 3% of all low-value services, they represent 34% of all low-value spending.( 7 ) Evidence-based recommendations for clinicians, including those from the multispecialty society Choosing Wisely campaign( 8 ) were developed in 2012 to reduce low-value services, including imaging. In fact, imaging studies account for 3 of the top 5 American Academy of Pediatrics list of low-value services.( 9 ) Using claims-based measures of low-value care, Chua, et al( 7 ) found that, in 2014, nearly 10% of commercially insured children received at least 1 low-value service. A follow-up study found a similar prevalence among publicly insured children. ( 10 ) Essential next questions to guide intervention development include where and how these services are distributed across settings. Such information is important for providers and payers to understand where resources and interventions should be focused in order to reduce these services. This is particularly salient to state Medicaid programs, as they insure nearly 40% of U.S. children.( 11 ) Thus, the objectives of our study were to estimate the prevalence of low-value imaging services among pediatric Medicaid beneficiaries across 5 conditions, to determine the distribution of these services across healthcare settings, and to estimate associated expenditures from these services.