Trends in Low-Value Care Among Children’s Hospitals Free

PHIS is an administrative database containing deidentified data with demographics, diagnoses, procedures, and daily billing information from 49 children’s hospitals. Participating hospitals are large children’s hospitals located in urban regions in 27 states and the District of Columbia. Data quality is ensured through a joint effort between the Children’s Hospital Association (Lenexa, Kansas) and participating hospitals.

Detailed methodology of the LVC Calculator’s development and full measure specifications were previously published.³  Briefly, the tool contains 30 LVC measures applicable to 2 patient cohorts:

1. ED cohort, defined as encounters resulting in discharge from the ED; and

2. Hospitalized cohort, defined as encounters for patients admitted to the hospital under inpatient or observation status.

In the hospitalized cohort, measured care includes that delivered as an inpatient and during the associated ED visit, because PHIS cannot distinguish between the 2 settings within the same facility. Some LVC measures are applicable to both cohorts, whereas others are appropriate for application in only 1.

Measures included in the LVC Calculator represent the delivery of medications, laboratory and imaging studies, and procedures delivered during clinical scenarios in which the evidence does not support their use. Measures were derived from quality measure and recommendation sets published by national organizations, or as part of previous research efforts.^4,23–26  The multidisciplinary stakeholder group responsible for the calculator’s development prioritized narrow measure definitions.³  This approach was intended to define LVC with high specificity, while minimizing misclassification of appropriate, or justified, care practices.^3,4  To do this, the LVC Calculator first applies a set of global exclusions to remove encounters with a high level of patient complexity or very severe illness. The tool then employs measure-specific exclusions to remove encounters with evidence of justification for a particular service.³  Measure definitions are shown in Supplemental Table 4.

The LVC Calculator is designed as both a research and quality improvement tool. It is used to measure characteristics of LVC delivery across PHIS hospitals; additionally, clinicians at participating hospitals can access data describing local LVC delivery patterns at any time and compare performance to peer hospitals.

We conducted a retrospective cohort study by applying the LVC Calculator to encounters between January 1, 2016, and December 31, 2022. Because measures were defined using International Classification of Diseases, 10th Revision, Clinical Modification, codes, 2016 was the first full year to which the LVC Calculator could be applied. Hospitals not contributing data for the entire study period were excluded.

Encounters were included for analysis if they met criteria for ≥1 of the LVC Calculator measures. Per the calculator’s parameters, encounters for patients aged >18 years and those with International Classification of Diseases, 10th Revision, codes indicating complex chronic conditions²⁷  or neurologic impairment²⁸  were excluded. Encounters with All Patient Refined Diagnosis-Related Groups (3M Healthcare) for extreme severity of illness and those with intensive care utilization (with the exception of neonatal intensive care-specific measures) were also excluded from the hospitalized cohort. Measure-specific exclusions were then applied to remove encounters with justification of low-value service delivery.³  Encounters lacking any of the above exclusion criteria served as the denominator for LVC analyses.

For all included measures, we calculated percentage of LVC delivery by year, defined as the number of eligible encounters with LVC (numerator) divided by the total number of eligible encounters (denominator). We used generalized estimating equations accounting for hospital clustering to analyze trends over time for individual measures, with year of encounter as a covariate; the resultant odds ratio and P value therefore reflect the significance of annual change over the study period.

After initial analyses revealed apparent trajectory changes for some measures coinciding with the onset of the severe acute respiratory syndrome coronavirus disease 2019 (COVID-19) pandemic in 2020, we performed an analysis comparing annual rates of change in the percentage of encounters with LVC delivery over 2 periods: (1) 2016 to 2019 (preceding the COVID-19 pandemic), and (2) 2020 to 2022 (onset of the pandemic to present). Rates of changes for these periods were compared with t tests to assess whether LVC patterns observed during and after the pandemic differed from those observed before the pandemic.

We additionally assessed observed changes by category of measure (medications, laboratory and imaging studies, and procedures). Finally, we ranked measures by the proportion and volume of encounters with LVC delivery at the conclusion of the study period.

After initial data review, 1 measure, group A streptococcus (GAS) testing for children <3 years of age for routine pharyngitis in the ED, was excluded from further analyses. The definition for this measure captures rapid antigen testing for GAS only; however, recent literature suggests a practice shift toward the use of molecular testing.^29,30  As such, apparent reductions in GAS testing may represent missed molecular testing (for which PHIS has not yet validated a coding approach).

Statistical analyses were performed with SAS, version 9.5 (SAS Institute, Cary, North Carolina), and P < .05 was considered statistically significant. This study was deemed not human subjects research by the Dartmouth College institutional review board.

There were 6 566 766 encounters eligible for ≥1 LVC Calculator measure among 43 hospitals contributing data for the entire study period (5 265 153 ED encounters; 1 301 613 encounters for hospitalizations). In both cohorts, encounters eligible for each measure were relatively stable from 2016 to 2019, but decreased markedly in 2020, as did all pediatric encounters in PHIS.³¹  By 2022, volume of eligible encounters approximated those observed at the start of the study period for most measures (Tables 1 and 2).

Of 21 measures applicable to the ED cohort, the proportion of eligible encounters in which LVC was delivered increased for 11 measures, suggesting declining performance over the study period (Table 1). LVC delivery decreased for 1 measure and was unchanged for 9 measures. Figure 1 shows trends for measures demonstrating a statistically significant change of ≥5% from 2016 to 2022. LVC increased by ≥5% for 3 measures: Head computed tomography (CT) for minor head injury (16.5% of eligible encounters in 2016 to 22.9% in 2022); head CT for first-time, generalized, afebrile seizure (12.7%–18.4%); and C-reactive protein and/or erythrocyte sedimentation rate for uncomplicated CAP (4.7% to 10.6%). LVC decreased by ≥5% for one measure: bronchodilators for bronchiolitis (22.3%–16.8%). By 2022, the proportion of encounters with LVC delivery was greatest for CT for minor head injury (22.9%) and afebrile seizure (18.4%); measures with the greatest volume of encounters with LVC delivery were chest radiography for asthma (10 468) and CT for minor head injury (8735) (Supplemental Table 5).

Table 2 demonstrates temporal trends in LVC delivery for the hospitalized cohort. Of 26 applicable measures, the proportion of eligible encounters in which LVC was delivered increased for 6 measures and decreased for 9 measures, whereas performance for 11 measures remained unchanged. Figure 2 demonstrates trends for measures with statistically significant changes of ≥5% from 2016 to 2022. LVC increased by ≥5% for 3 measures: Electrolyte testing for simple febrile seizure (27%–38.4%), C-reactive protein and/or erythrocyte sedimentation rate for CAP (23.2%–37.9%), and head CT for first-time generalized afebrile seizure (6.9%–13.3%). LVC decreased by ≥5% for five measures: Broad-spectrum antibiotic therapy for CAP (60.3%–48.3%), acid suppression medications administered to children hospitalized on the pediatric ward (54.9%–49.4%) and those in the NICU (15.8%–8.6%), peripherally-inserted central catheter placement for bone and joint infections (15.7%–6.1%), and bronchodilators for bronchiolitis (37.4%–30.9%). By 2022, the proportion of encounters with LVC delivery was greatest for acid suppression therapy for infants with gastroesophageal reflux (49.4%) hospitalized on the pediatric ward and broad-spectrum antibiotic therapy (48.3%); measures with the greatest volume of encounters with LVC delivery were chest radiography for asthma (6963) and bronchodilators for bronchiolitis (6868) (Supplemental Table 5).

Table 3 compares annual rates of change by measure from 2016 to 2019 and from 2020 to 2022. No statistically significant difference in annual rates of change was noted between these 2 time periods for any measure.

Supplemental Table 6 shows performance by measure category. LVC was either unchanged or decreased for nearly all medication measures in both cohorts; the exception was steroids for bronchiolitis, which increased in both cohorts over the study period (Tables 1 and 2). For the 3 procedure measures (only applicable to the inpatient cohort), LVC was also unchanged or decreased. For all 5 laboratory measures, LVC was either unchanged or increased in both cohorts. For the 9 imaging measures, LVC was also unchanged or increased for nearly all measures in both cohorts, with the exception of chest radiography for bronchiolitis in the inpatient cohort, which decreased (Table 2).

Our study identified varying performance trajectories across a set of hospital-based LVC measures from 2016 to 2022, with differences identified by measure category. LVC was unchanged or decreased for most medication and procedural measures; conversely, LVC was either unchanged or increased for all laboratory measures and nearly all imaging measures. Among all 29 measures, the majority demonstrated relative performance stagnation over the study period, with many demonstrating persistently high rates of LVC delivery in 2022.

This study adds to existing literature by exploring a broad set of hospital-based pediatric LVC measures and providing recent longitudinal data for several previously studied measures. We found ongoing improvement in LVC delivery for some services, such as bronchodilators for bronchiolitis. This continued improvement may reflect the robust evidence base,³²  dissemination of national recommendations,³³  and numerous quality improvement initiatives dedicated to reducing LVC for this common diagnosis.³⁴  However, we also identified reversal of previous improvement trends for some measures, including chest radiography for asthma,²  head CT for minor head injury,³⁵  and corticosteroids for bronchiolitis.¹⁷  These findings may suggest challenges in sustainability of deimplementation efforts; additionally, they support a need for consistent LVC measurement to facilitate timely identification of such shifts. Our observation that LVC failed to improve for a majority of included measures mirrors results from several studies focused on adult populations, which have found minimal improvements in LVC over time despite an intensified awareness of the harms associated with these services.^36–38 

The differential patterns observed between measures associated with the diagnostic process (laboratory and imaging studies, for which LVC was generally unchanged or increasing) and those associated with treatment (medications and procedures, for which LVC was unchanged or decreasing) is notable. Although reductions in low-value medications and procedures are key successes, efforts are needed to deimplement services further upstream in the diagnostic process. Such attention may reduce misdiagnosis and overdiagnosis³⁹  and prevent cascades of LVC, defined as medical services that directly follow from an initial low-value service.⁴⁰  For example, avoidance of chest radiography for bronchiolitis not only reduces direct costs and harm from radiation, but may also prevent a misdiagnosis of bacterial pneumonia on the basis of nonspecific imaging findings (eg, a focal opacity which may be consistent with atelectasis) and therefore avoid a course of antibiotics and associated harms. Our findings highlight the importance of continued efforts to measure the downstream impact of diagnostic services and focus on diagnostic stewardship.⁴¹ 

Our results raise some questions surrounding the potential influence of the COVID-19 pandemic on pediatric LVC. Literature supports a marked reduction in pediatric health care services at the start of the pandemic accompanied by shifts in common disease patterns,⁴²  yet little is known about pandemic-related changes in health care value. For some LVC Calculator services, such as head CT for minor head injury (ED cohort) and for generalized seizure (both cohorts), inflammatory marker testing for uncomplicated CAP (both cohorts), and electrolyte testing for febrile seizures (hospitalized cohort), a large increase in LVC was observed after the onset of the COVID-19 pandemic. It is possible that changes introduced by the pandemic may have directly altered LVC patterns. For example, providers may have felt more compelled to obtain definitive diagnostic information to avoid longer observation periods in the health care setting and reduce likelihood of follow-up care needs. Alternatively, alterations in patient acuity may have influenced these trends. For example, children seeking care for minor head injury during the pandemic may have had more severe symptomatology than those presenting during other periods of our study, because caregivers may have attempted to minimize health care contact during the pandemic due to infectious risks. The trajectories of these measures in the postpandemic period have varied, with some increases continuing and others tempering. Annual rates of change do not yet demonstrate significant differences for any services between the prepandemic period (2016–2019) and the study period inclusive of the pandemic and postpandemic years (2020–2022). Further measurement, particularly of services that are continuing to increase, and assessment of drivers will be needed to elucidate the impact of the pandemic on LVC.

Given lack of improvement in LVC delivery for many measures over this study period, a structured approach is needed to identify and prioritize services for which deimplementation efforts will be most impactful. The LVC Calculator may assist with achieving this aim by offering continuous data describing multiple aspects of LVC measurement including prevalence, cost, and temporal trends. We identified temporal increases in some measures, such as CT for minor head injury (ED cohort) and for first-time generalized seizure (both cohorts) for which LVC has been shown to be particularly costly³  and harmful,^43–45  suggesting these services are important areas for prioritized deimplementation. Targeted deimplementation may also benefit measures demonstrating stagnant performance where LVC remains prevalent. For example, in the ED cohort, chest radiography remained unchanged through the study period, but was delivered in >10 000 encounters in 2022 (nearly 17% of those eligible); in the hospitalized cohort, blood cultures for CAP also remained unchanged, but were obtained in >1500 encounters (nearly 40% of those eligible). Our previous work also found these services to be relatively costly and to impact a large number of children.³  For such measures, comparing current performance to achievable benchmarks of care^22,46  will identify realistic improvement potential and offer improvement targets. Further work is needed to systematically assess other key areas of LVC impact, including perceived degree of harm associated with each service, potential for cascades of care, and disparities that may impact health equity; additionally, future efforts should focus on further characterizing drivers and outcomes associated with this care.

Our study has important limitations. First, our definition of LVC relies on administrative data. Although the measure definitions in the LVC Calculator are intentionally narrow, intending to identify services that are truly low-value, it is possible that some services were misclassified as low-value. Conversely, it is possible that the measure definitions may underestimate the burden of LVC. Although the LVC Calculator’s design is intended to exclude children with severe illness, it is possible that variations in patient characteristics, including acuity, throughout the study period not captured by our exclusion criteria may have influenced LVC receipt. We did not perform statistical analyses describing the significance of short-term changes in LVC delivery during the pandemic; future analyses focused specifically on this period may further elucidate the direct impact of the pandemic on LVC. We did not cluster our analysis at the patient level and are therefore unable to assess the influence of repeated low-value service delivery to an individual patient on our results. Finally, our analysis includes only data from children’s hospitals participating in PHIS, and our findings may not be generalizable to other settings.

This longitudinal assessment of 29 pediatric hospital-based LVC measures demonstrated varying performance trends by measure from 2016 to 2022, with a persistently high prevalence of LVC observed for many measures at the conclusion of the study period. Performance trends were less favorable for diagnostic measures (imaging and laboratory studies) than therapeutic measures (medications and procedures). Given the imperative to reduce LVC delivery and realistic limitations in deimplementation resources, these data may inform prioritization of deimplementation efforts.

CAP

community-acquired pneumonia

COVID-19

coronavirus disease 2019

CT

computed tomography

ED

emergency department

GAS

group A streptococcus

LVC

low-value care

PHIS

Pediatric Health Information System