Trends in Capability of Hospitals to Provide Definitive Acute Care for Children: 2008 to 2016

We conducted a national longitudinal study of EDs and ED visits made by children from 2008 to 2016 by using the Healthcare Cost and Utilization Project Nationwide Emergency Department Sample (NEDS) database, which includes all ED visits to an annually varying 20% stratified sample of all hospital-owned EDs nationwide. Although the sampling strategy of the NEDS from 2008 to 2016 was weighted to the entire United States, the number of states contributing data has changed over time, from 28 states in 2008 to 37 states in 2016.¹⁵  The NEDS provides a broad view of ED and acute inpatient encounters through national estimates of hospital- and visit-level characteristics.^15–17 

For visit-level analyses, we included all ED visits by children <15 years old in keeping with previous pediatric capability research.⁸  We excluded psychiatric visits because transfer and capability patterns for these conditions differ substantially from all others.⁸  Visits with missing disposition (1.6% of visits) were dropped from visit-level analyses but were included in counts of total visits per ED. For hospital-level analyses, we included all institutions except those without ED visits by children.

The primary outcome was the Hospital Capability Index⁷  (HCI), which is used to measure the degree to which a hospital provides inpatient care for patients not discharged from the ED. It is computed for each condition as the number of hospitalizations divided by transfers plus hospitalizations; a hospital’s overall HCI score is the mean of the values for all conditions and is thus unweighted by condition prevalence.^7,8  The HCI score varies from 0 to 1. A facility with an HCI score of 0 does not provide pediatric inpatient care (always transfers instead of hospitalizes) for any condition. The HCI score was recalculated each year for every sampled hospital. A visit’s assigned condition was the top level Clinical Classifications Software code corresponding to the first-listed diagnosis for the visit.¹⁸  Each year we dropped from the calculation conditions with fewer than 500 ED visits nationally due to extreme rarity (eg, gangrene) because they would unduly influence the HCI score.

To evaluate the likelihood that definitive emergency care was provided, we used a secondary measure of capability, the ED transfer rate, calculated as pediatric transfers divided by all pediatric ED visits. The key differences in how HCI score and ED transfer rate are calculated are that HCI score is unweighted to condition prevalence, whereas ED transfer rate is influenced by condition prevalence, and the HCI does not include ED discharges in the denominator, whereas the ED transfer rate does.

The main predictor variable was the annual ED volume of children,^4,19,20  categorized as low (<1800 pediatric visits per year), medium (1800–4999), medium-high (5000–9999), high (≥10 000), or primarily pediatric.¹  Centers were primarily pediatric if ≥70% of all ED visits were made by children younger than age 15.¹⁶  Pediatric volume was recategorized for each hospital yearly. Although we did not prespecify that primarily pediatric hospitals meet a volume threshold, all had >10 000 pediatric visits each year.

Other hospital-level covariates included the proportion of patients with Medicaid as the primary payer and the urbanicity of the hospital county: large metropolitan (≥1 million residents), small metropolitan (50 000–999 999 residents), micropolitan (10 000–49 999 residents), or rural (<10 000 residents), defined by Urban Influence Codes.²¹  We treated a nonspecific urbanicity designation as missing (3.6% of unweighted hospital-year pairs).

Visit-level covariates included patient age (<1, 1–4, 5–7, 8–11, or 12–14 years), patient sex, quartile of median income of the patient’s ZIP code, payer (Medicaid or non-Medicaid), and patient’s home urbanicity.

HCI and ED transfer rate trends by annual volume and urbanicity were plotted.²²  We calculated the magnitude of changes in HCI score and ED transfer rates between 2008 and 2016 using differences in medians and assessed statistical significance using Wilcoxon rank-sum tests. We performed a preplanned subanalysis to assess whether uncommon conditions drove HCI trends by plotting HCI trends over time for the 20 most common conditions. To assess the association of capability to care for all conditions with capability to care for common conditions, we used a weighted Pearson correlation.

To assess how patients’ access to pediatric-capable hospitals changed between 2008 and 2016, we defined 4 capability categories on the basis of the 2008 HCI quartile cutoffs, thus treating 2008 as a baseline. Hospitals were reassigned to 1 of these capability categories each year. We displayed trends in the number of hospitals and ED visits in each capability category. We then examined to what extent children of differing urbanicity have access to hospitals outside the lowest capability category.

To investigate potential reasons hospitals may decrease in capability over time, we first determined the annual proportion of eligible hospitals that moved into a higher or lower capability category in the next sampled observation. Eligible hospitals were those that were included in a subsequent year of data collection and could change category upward or downward (eg, hospitals in the bottom category could not switch to a lower category). We plotted annual counts of changes in category.

We then constructed Kaplan-Meier curves of time to switching into the lowest capability category, stratified by initial category. Hospitals were censored at the last observed year. A hospital was at-risk from the first year it was included in the NEDS until the event or censoring. We created a Cox proportional hazards model predicting time to switching, using covariates we hypothesized would impact capability degradation. These covariates were measured the first year a hospital was sampled and included capability category, pediatric volume, hospital urbanicity, primary payer, and region. Hospital volume or crowding may have an impact on hospitals’ decisions to hospitalize children,^14,23  so we included 3 time-varying measures of change in hospital volume from entry to switching to the lowest capability category or censoring. These included change in overall ED volume, pediatric inpatient volume, and adult inpatient volume, each categorized as decreased or unchanged, increased ≤20%, increased >20 to ≤40%, or increased >40%. No primarily pediatric hospitals moved into the lowest capability category, so they were not included in this analysis.

All analyses were performed using survey-weighted counts, proportions, statistical tests, regression models, and figures (eg, weighted boxplots and Kaplan-Meier curves) in accordance with Healthcare Cost and Utilization Project best practices.²⁴  We used 95% confidence intervals (CIs) and defined statistical significance as a 2-sided P value <.05. The Boston Children’s Hospital Institutional Review Board considered this study exempt from review.

We analyzed 3020 unique hospitals (945–980 unweighted hospital observations per year, weighted to represent 4548–4864 hospitals). After sample weighting, the data represented 207.4 million ED visits among children <15 years. We excluded 2.4 million (1.15%) for a primary mental health diagnosis and therefore analyzed 205.0 million weighted ED visits. Unique EDs in the NEDS had a mean of 2.9 data years available (SD: 1.7; range: 1–9).

Hospital characteristics by pediatric volume are displayed in Table 1. High-volume and primarily pediatric hospitals disproportionately exist in metropolitan areas and treat a larger proportion of Medicaid patients compared with lower-volume EDs. EDs with low pediatric volume were the most common ED type (38.5%), and primarily pediatric EDs were the least common (0.8%, Supplemental Fig 4). Although 33.0% of EDs are in large metropolitan areas, 53.8% of visits occurred at such EDs (Supplemental Fig 4). Payer mix changed over time. The median (interquartile range [IQR]) proportion of children with Medicaid increased from 39.8% (IQR: 25.3–51.5) in 2008 to 48.8% (IQR: 36.1–61.5) in 2016 (P < .001). Demographic features of ED visits appear in Supplemental Table 3. Conditions used in the calculation of HCI scores exhibited little variation by year. Of the 246 Clinical Classifications Software codes defining conditions, a median 86 (IQR: 82–89) were excluded from the HCI calculation for not meeting the prespecified inclusion criterion of 500 visits in a year, and 50 conditions were excluded in all 9 years.

Across all hospitals, the median HCI score was 0.06 (IQR: 0.01–0.17) in 2008 and 0.02 (IQR: 0.00–0.09) in 2016 (P < .001). This indicates that for the median hospital in 2008, averaged across conditions, 6% of children who could not be discharged from the ED were hospitalized at the index hospital. Median HCI score was higher in hospitals in categories of higher pediatric volume. Provision of definitive care as measured by HCI score decreased over time in all categories of pediatric volume except primarily pediatric hospitals (P < .001 for each 2016 vs 2008 comparison, Fig 1). Median HCI score decreased between 2008 and 2016 by 0.02 in low-volume centers, by 0.04 in medium-volume centers, by 0.09 in medium-high–volume centers, and by 0.15 in high-volume, nonprimarily pediatric centers, whereas the median HCI score for primarily pediatric centers did not change significantly (increase of 0.03, P = .12).

Total childhood ED visits nationally increased from 21.8 million in 2008 to 24.1 million in 2016, but the difference did not reach statistical significance (increase of 11.0%; 95% CI: −3.2 to 25.1). However, over this same time period, total ED transfers increased 27.7% (95% CI: 9.8 to 45.6), from 234 230 to 299 171. Across all hospitals, the ED transfer rate increased between 2008 and 2016 from a hospital median of 1.3% (IQR: 0.6–2.0) to 1.8% (IQR: 0.9–2.8, P < .001). ED transfer rates increased in low- (risk difference: +0.45%), medium- (risk difference: +0.59%), and medium-high–volume EDs (risk difference: +0.53%) (P < .001 for each comparison). ED transfer rates did not change significantly during the study period in high-volume (P = .06) or primarily pediatric EDs (P = .90).

Definitive care decreased over time regardless of urbanicity (Supplemental Fig 5, P < .001 for each 2016 vs 2008 HCI score comparison), but median HCI score was higher in more urban areas than in more rural areas. Overall transfers increased during 2008 to 2016 in all urbanicity categories (P < .001 for each 2016 vs 2008 comparison).

The most common conditions evaluated among children in the ED were upper respiratory infections, otitis media, and superficial injuries or contusions. The 20 most common conditions in children comprised 72.7% of ED visits (Supplemental Table 4). The national median HCI score among visits for common conditions decreased from 0.20 (IQR: 0.03–0.44) in 2008 to 0.04 (IQR: 0.00–0.28) in 2016 (P < .001). Capability to care for common conditions was strongly correlated with overall HCI score (weighted Pearson r: 0.92). Pediatric capability decreased significantly for common conditions, except in primarily pediatric hospitals; ED transfer rates increased, except in high-volume and primarily pediatric hospitals (Supplemental Fig 6).

In 2008, the cutoff defining the lowest capability category was an HCI score of 0.013. Between 2008 and 2016, the proportion of EDs below this threshold increased from 25.0% (95% CI: 22.5% to 27.8%) to 49.0% (95% CI: 46.1% to 51.9%). ED visits to hospitals below this threshold increased from 11.5% (95% CI: 9.5% to 13.8%) to 22.0% (95% CI: 18.7% to 25.6%) (Fig 2).

Changes in access to hospitals outside the lowest capability category differed by urbanicity. Visits specifically among rural children to hospitals in the lowest capability category increased from 12.6% (0.2 out of 1.7 million) in 2008 to 39.1% (0.6 out of 1.6 million) in 2016. By contrast, visits to hospitals in the lowest capability category among children in central urban areas increased from 11.7% (0.7 out of 5.7 million) to 19.1% (1.5 out of 8.0 million).

On the basis of the capability category thresholds defined in 2008, 912 out of 3296 (27.7%) hospitals switched to a lower category in a subsequent assessment for that hospital, whereas 297 out of 3231 (9.2%) switched to a higher category (Fig 3). In 2015, 211 out of 868 (24.3%) switched downward over the subsequent year, and 72 out of 1108 (6.5%) switched upward.

Increasing pediatric hospitalization volume and higher capability category were associated with a lower likelihood of switching to the lowest capability category (Table 2, Supplemental Fig 7).

Across acute care hospitals nationally, provision of definitive hospital care to children is declining, and ED transfers of children are increasing. Definitive care was less likely to occur at hospitals with low pediatric volume and at rural hospitals but declined over time regardless of ED volume or urbanicity, except at primarily pediatric facilities. Similarly, ED transfers increased over time by 27%, with increases in all volume and urbanicity categories, except at primarily pediatric institutions, and the sharpest increases occurred among low-volume and rural hospitals. Capability decreases occurred across all conditions, as well as the 20 most common conditions, suggesting that transfers of uncommon conditions are not driving trends in definitive care. Using the lowest quartile of the 2008 capability index as a benchmark, the number of hospitals below this threshold nearly doubled between 2008 and 2016, resulting in an additional 2.8 million pediatric visits to these EDs annually. Together, these trends indicate that the provision of definitive ED and hospital care is becoming less common in most institutions and transfer to another hospital is becoming increasingly common.

Availability of resources to care for children in the ED has been improving in hospitals nationally, including improved equipment, staff, training, and protocols.^1,5,6,25  However, this focus on ED resource availability has occurred even as we observe a decrease in hospitals providing definitive care. The trends we observed are concordant with previous capability work, although our HCI estimates are lower than the pediatric HCI scores of 0.19, 0.08, 0.18, and 0.16 observed in California, Florida, Massachusetts, and New York, respectively.⁹  The differences in values suggest that the national performance is lower than in these 4 states. Decreasing capability may reflect regionalization. Regionalizing care has been shown to help decrease costs and improve quality for some services. Benefits of regionalization include improved mortality among infants with very low birth weight at high-volume neonatal intensive care centers,^10,26  improved appendicitis and pyloric stenosis outcomes with specialized pediatric surgical care,²⁷  decreased disability and mortality after serious injury for those receiving care in level 1 trauma centers,^28,29  and improved postcardiac arrest survival after regionalization of cardiac care.³⁰  The benefits of regionalization of emergency care in general are less clear.³¹  Regionalization may have the unintended effect of decreasing access to care in areas served only by smaller hospitals or geographically distant from a referral center.^11,30,32  Regionalization may also increase unnecessary transfers and inconvenience to children with lower-acuity conditions as well as increase costs to the health care system.³³ 

The outcomes associated with declines in hospital-based provision of definitive care are unclear. At a minimum, given the large increases in visits to hospitals in the lowest capability category, an increasing proportion of the country now lacks easy access to a hospital that is likely to provide definitive hospital care. This is particularly true in rural areas, where pediatric volume and capability are lower than in nonrural areas. In combination with overall increases in hospital closures in rural areas,^34,35  rural children are most vulnerable to declining access to high-capability care.³²  Continued declines in capability and increases in transfers may also lead to a cycle of decreasing clinician experience with pediatric care in general, further driving transfer rates up.^31,33,36,37  Beyond that, increases in transfers generally delay care, which may be important in serious conditions, including appendicitis, trauma, and others.^38–41  Whether a decline in definitive care capability impacts the quality of initial stabilization of patients at referring hospitals, leads to worse outcomes, or alters system capacity is unknown but is a crucial future area of study.³⁶ 

The decline in provision of definitive hospital-based acute care likely has multiple causes. First, declining pediatric hospitalizations may contribute to decreased experience with hospital care as well as financial disincentives for maintaining pediatric inpatient capacity.³  The fixed costs of maintaining pediatric resources such as imaging, equipment, and trained staff relative to the decreasing volume may become more difficult to justify. Second, increasing illness severity or decreasing low-acuity visits among children could warrant increases in transfers, in keeping with evidence suggesting increases in the incidence of serious emergency conditions⁴²  and complex comorbidities in children.^43–45  However, available evidence does not suggest shifts of low-acuity pediatric visits away from EDs.^46,47  Third, crowding within existing inpatient units might lead to increases in transfers for capacity reasons. However, our model indicates the largest increases in inpatient volume are least associated with decreasing capability, and ED visit volume is not associated with a hospital’s capability over time. Fourth, increases in discretionary transfers at a parent’s or primary care physician’s request may lead to apparent decreased capability, although there is no available evidence of such trends. Finally, financial incentives may encourage transfer because increasing numbers of children are insured by Medicaid, which has been associated with an increased likelihood of transfer.^20,48 

Regardless of the causes, care is becoming more concentrated, and fewer patients are receiving definitive care at their first choice of hospital. As pediatric patient complexity increases, transfers will likely continue to increase, with a possible consequence of crowding at regional centers.

This study has several limitations. First, there are few primarily pediatric centers in the United States, which reduces the precision of their capability and transfer estimates. This may have led to outlier HCI estimates among primarily pediatric hospitals in 2012. Precisely assessing these centers is important because the high volumes, ED and inpatient capacities, subspecialty access, resource availability, and referral environment make them major recipients of regional transfers.^16,19,49  Second, we were unable to distinguish high-volume general hospitals with dedicated pediatric ED, inpatient, specialty, and critical care services (which may function similarly to primarily pediatric hospitals) from those with high volume but fewer dedicated resources (which are a majority of high-volume hospitals¹ ). This may have accounted for some of the variation in HCI scores observed across the high-volume category because these 2 types of hospitals differ in capability. Finally, because the number of years sampled per hospital varied, we could not know the precise year some hospitals moved into the lowest capability category in the survival analysis.

Across the nation, hospitals are consistently declining in their capability to provide definitive hospital-based acute care to children and are increasingly likely to transfer pediatric patients. The number of pediatric visits to hospitals classified in the lowest capability category has more than doubled since 2008, with rural children most affected. Assessment of the impact of capability on pediatric acute care outcomes and investment in community pediatric care may be needed to maintain access to definitive care for children.

CI

confidence interval

ED

emergency department

HCI

Hospital Capability Index

IQR

interquartile range

NEDS

Nationwide Emergency Department Sample